Lokenath Debnath
Nonlinear Partial Differential Equations for Scientists and Engineers Third Edition
Lokenath Debnath Department of Mathematics University of Texas, Pan American 1201 W. University Drive Edinburg, TX, 78539 USA
[email protected]
ISBN 978-0-8176-8264-4 e-ISBN 978-0-8176-8265-1 DOI 10.1007/978-0-8176-8265-1 Springer New York Dordrecht Heidelberg London Library of Congress Control Number: 2011940003 Mathematics Subject Classification (2010): 00A06, 00A69, 33Exx, 33E12, 34B05, 34B24, 34B27, 34B60, 34G20, 34G25, 35A15, 35A21, 35A22, 35A25, 35C05, 35C15, 35Dxx, 35E05, 35E15, 35Fxx, 35F05, 35F10, 35F15, 35F20, 35F25, 35G10, 35G20, 35G25, 35J05, 35J10, 35J15, 35J20, 35K05, 35K10, 35K15, 35K55, 35K60, 35L05, 35L10, 35L15, 35L20, 35L25, 35L30, 35L60, 35L65, 35L67, 35L70, 35Q30, 35Q35, 35Q40, 35Q51, 35Q53, 35Q55, 35Q60, 42A38, 44A10, 44A35, 49J40, 49Lxx, 58E30, 58E50, 65L15, 65M25, 65M30, 65R10, 70H05, 70H06, 70H09, 70H20, 70H25, 70H30, 76Bxx, 76B07, 76B15, 76B25, 76B55, 76B60, 76B65, 76D05, 76D33, 76D45, 76E30, 76M30, 76R50, 78M30, 81Q05, 81Q10 © Springer Science+Business Media, LLC 2012 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper Springer is part of Springer Science+Business Media (www.birkhauser-science.com)
TO MY MOTHER with love, gratitude, and admiration
True Laws of Nature cannot be linear. Albert Einstein . . . the progress of physics will to a large extent depend on the progress of nonlinear mathematics, of methods to solve nonlinear equations . . . and therefore we can learn by comparing different nonlinear problems. Werner Heisenberg Our present analytical methods seem unsuitable for the solution of the important problems arising in connection with nonlinear partial differential equations and, in fact, with virtually all types of nonlinear problems in pure mathematics. The truth of this statement is particularly striking in the field of fluid dynamics. . . . John Von Neumann However varied may be the imagination of man, nature is a thousand times richer, . . . Each of the theories of physics . . . presents (partial differential) equations under a new aspect . . . without these theories, we should not know partial differential equations. Henri Poincaré Since a general solution must be judged impossible from want of analysis, we must be content with the knowledge of some special cases, and that all the more, since the development of various cases seems to be the only way to bringing us at last to a more perfect knowledge. Leonard Euler . . . as Sir Cyril Hinshelwood has observed . . . fluid dynamicists were divided into hydraulic engineers who observed things that could not be explained and mathematicians who explained things that could not be observed. James Lighthill
Preface to the Third Edition
A teacher can never truly teach unless he is still learning himself. A lamp can never light another lamp unless it continues to burn its own flame. The teacher who has come to the end of his subject, who has no living traffic with his knowledge but merely repeats his lessons to his students, can only load their minds; he cannot quicken them. Rabindranath Tagore An Indian Poet 1913 Nobel Prize Winner for Literature The previous two editions of my book were very well received and used as a senior undergraduate or graduate-level text and research reference in the United States and abroad for many years. We received many comments and suggestions from many students, faculty, and researchers around the world. These comments and criticisms have been very helpful, beneficial, and encouraging. This third edition is the result of the input. Another reason for adding this third edition to the literature is the fact that there have been major discoveries of new ideas, results and methods for the solutions of nonlinear partial differential equations in the second half of the twentieth century. It is becoming even more desirable for mathematicians, scientists, and engineers to pursue study and research on these topics. So what has changed, and will continue to change, is the nature of the topics that are of interest in mathematics, applied mathematics, physics, and engineering, the evolution of books such as this one is a history of these shifting concerns. This new and revised edition preserves the basic content and style of the second edition published in 2005. As with the previous editions, this book has been revised primarily as a comprehensive text for senior undergraduates or beginning graduate students and a research reference for professionals in mathematics, engineering, and other applied sciences. The main goal of the book is to develop required analytical skills on the part of the reader, rather than to focus on the importance of more abstract formulation, with full mathematical rigor. Indeed, our major emphasis is to
viii
Preface to the Third Edition
provide an accessible working knowledge of the analytical and numerical methods with proofs required in mathematics, applied mathematics, physics, and engineering. In general, changes have been made to modernize the contents and to improve the exposition and clarity of the previous edition, to include additional topics, comments, and observations, to add many examples of applications and exercises, and in some cases to entirely rewrite and reorganize many sections. There is plenty of material in the book for a year-long course or seminar. Some of the material need not be covered in a course work and can be left for the readers to study on their own in order to prepare them for further study and research. This edition contains a collection of over 1000 worked examples and exercises with answers and hints to selected exercises. Some of the major changes and additions include the following: 1. Many sections of almost all chapters have been revised and expanded to modernize the contents. We have also taken advantage of this new edition to correct typographical errors and to include several new figures for a clear understanding of physical explanations. 2. Several nonlinear models including the Camassa–Holm (CH) equation, the Degasperis–Procesi (DP) equation, and the Toda lattice equation (TLE) have been presented with their physical significance in Chapter 2. Included are also new sections on the small-amplitude gravity-capillary waves on water of finite and infinite depth, the energy equation and energy flux. 3. A new section on the Lorenz nonlinear system, the Lorenz attractor, and deterministic chaos has been added in Chapter 6. 4. Included is a new section on the Camassa–Holm equation, the Degasperis– Procesi equation, and the Euler–Poincaré (EP) equation in Chapter 9 to describe the wave breaking (singular) phenomena. A new section on the derivation of the KdV equation for the gravity-capillary wave, the gravity-capillary solitary wave solutions, and the two-dimensional periodic flow in an inviscid, incompressible fluid with constant vorticity has been added. Special attention is given to both analytical and computational solutions of these problems with physical significance. 5. A new example describing nonlinear quasi-harmonic waves and modulational instability has been added in Chapter 10. 6. The nonlinear lattices and the Toda lattice equation have been treated in some detail at the end of Chapter 11. 7. All tables of Fourier transforms, Fourier sine and cosine transforms, Laplace transforms, Hankel transforms, and finite Hankel transforms have been revised and expanded so that they become more useful for the study of partial differential equations. 8. In order to make the book self-contained, two new appendices on some special functions and their basic properties, Fourier series, generalized functions, Fourier and Laplace transforms have been added. Special attention has been given to algebraic and analytical properties of the Fourier and Laplace convolutions with applications.
Preface to the Third Edition
ix
9. The whole section on Answers and Hints to Selected Exercises has been revised and expanded to provide additional help to students. 10. The entire bibliography has been revised and expanded to include new and current research papers and books so that it can stimulate new interest in future study and research. 11. The Index has been revised and reorganized to make it more useful for the reader. Some of the highlights in this edition include the following: •
•
•
•
•
•
The book offers a detailed and clear explanation of every concept and method that is introduced, accompanied by carefully selected worked examples, with special emphasis given to those topics in which students experience difficulty. A wide variety of modern examples of applications has been selected from areas of partial differential equations, quantum mechanics, fluid dynamics, solid mechanics, calculus of variations, linear and nonlinear wave propagation, telecommunication, soliton dynamics, and nonlinear stability analysis. The book is organized with sufficient flexibility to enable instructors to select chapters appropriate for courses of differing lengths, emphases, and levels of difficulty as chapters are significantly independent of each other. A wide spectrum of exercises has been carefully chosen and included at the end of each chapter so the reader may further develop both rigorous skills in the theory and applications of partial differential equations and a deeper insight into the subject. Many new research papers and standard books have been added to the bibliography to stimulate new interest in future study and research. The Index of the book has also been completely revised in order to include a wide variety of topics. The book provides information that puts the reader at the forefront of current research.
With the improvements and many challenging worked out problems and exercises, we hope this edition will continue to be a useful textbook for students as well as a research reference for professionals in mathematics, applied mathematics, physics, and engineering. It is my pleasure to express our grateful thanks to many friends, colleagues, and students around the world who offered their suggestions and help at various stages of the preparation of the book. Special thanks to Mrs. Veronica Chavarria for drawing some figures, typing the manuscript with constant changes and revisions. In spite of the best efforts of everyone involved, some typographical errors doubtless remain. Finally, we wish to express our special thanks to Mr. Tom Grasso, Senior Editor, and the staff of Birkhäuser, Boston, for their help and cooperation. I also wish to thank Mr. Donatas Akmanaviˇcius and his staff for their meticulous job in preparing the final revised manuscript for printing the third edition. I am indebted to my wife, Sadhana, for her understanding and tolerance while the third edition was being written. Edinburgh, Texas
Lokenath Debnath
Preface to the Second Edition
This is a revised and expanded version of the first edition, published in 1997. The first edition was well received and used as a graduate level text and research reference in the United States and abroad for the last several years. I have received many criticisms and suggestions from graduate students and faculty members around the world. Their suggestions for improvement have been very helpful, beneficial, and encouraging. Most of the changes have been made in response to that input. However, an attempt has been made to preserve the character of the first edition. I believe that this new edition will remain a major source of linear and nonlinear partial differential equations and provide a useful working knowledge of the most important analytical methods of the solutions of the equations. Finding and interpreting the solutions of these equations is a central part of the modern applied mathematics, and a thorough understanding of partial differential equations is essential for mathematicians, scientists, and engineers. The main emphasis of the book is on the development of the required analytical skills on the part of the reader, rather than the importance of more abstract formulation with full mathematical rigor. However, because the study of partial differential equations is a subject at the forefront of current research, I have made an effort to include many new ideas, remarkable observations, and new evolution equations as further research topics for the ambitious reader to pursue. I have taken advantage of this new edition to add some recent exciting developments of the subject, to update the bibliography and correct typographical errors, to include many new topics, examples, exercises, comments, and observations, and, in some cases, to entirely rewrite whole sections. The most significant difference from the first edition is the inclusion of many new sections, such as those on Sturm– Liouville (SL) systems and some major general results including eigenvalues, eigenfunctions, and completeness of SL system, energy integrals and higher dimensional wave and diffusion equations in different coordinate systems, solutions of fractional partial differential equations with new examples of applications, the Euler–Lagrange variational principle and the Hamilton variational principle with important examples of applications, and the Hamilton–Jacobi equation and its applications. Included also are the Euler equation and the continuity equation, which provide the fundamental basis of the study of modern theories of water waves, Stokes’ analysis of nonlinear
xii
Preface to the Second Edition
finite amplitude water waves, Whitham’s equation, peaking and breaking of water waves, and conservation laws of the Whitham equation. This edition also contains some recent unexpected results and discoveries including a new class of strongly dispersive nonlinear evolution equations and compactons, new intrinsic localized modes in anharmonic crystals, and the derivation of the Korteweg–de Vries (KdV) equation, Kadomtsev–Petviashvili (KP) equation, Boussinesq equation, axisymmetric KdV equation, and Johnson concentric equation derived from the asymptotic expansion of the nonlinear water wave equations. As an example of an application of compactons, the solution of nonlinear vibration of an anharmonic mass–spring system is presented. Included are the existence of peakon (singular) solutions of a new strongly nonlinear model in shallow water described by Camassa and Holm equation, and the Harry Dym equation, which arises as a generalization of the class of isospectral flows of the Schrödinger operator. Furthermore, asymptotic expansions and the method of multiple scales, formal derivations of the nonlinear Schrödinger equation, and the Davey–Stewartson nonlinear evolution equations with several conservation laws have been added to this edition. Several short tables of the Fourier, Laplace, and Hankel transforms are provided in Chapter 13 for additional help to the reader. A systematic mathematical treatment of linear and nonlinear partial differential equations is presented in the most straightforward manner, with worked examples and simple cases carefully explained with physical significance. Many and varied useful aspects, relevant proofs and calculations, and additional examples are provided in the numerous exercises at the end of each chapter. This edition contains over 600 worked examples and exercises with answers and hints to selected exercises, accompanied by original reference sources which include research papers and other texts. There is plenty of material in the book for a year-long course. Some of the material need not be covered in a course work and can be left for the readers to study on their own in order to prepare them for further study and research. It is my pleasure to express my grateful thanks to the many friends and colleagues around the world who offered their suggestions and generous help at various stages of the preparation of this book. I offer my special thanks to Dr. Andras Balogh for drawing all figures, to Dr. Dambaru Bhatta for proofreading the whole book, and to Ms. Veronica Martinez for typing the manuscript with constant changes and revisions. In spite of the best efforts of everyone involved, some typographical errors doubtless remain. Finally, I wish to express my special thanks to Mr. Tom Grasso and the staff of Birkhäuser, Boston, for their help and cooperation. I am deeply indebted to my wife, Sadhana, for her understanding and tolerance while the second edition was being written. University of Texas–Pan America
Lokenath Debnath
Preface to the First Edition
Overview Historically, partial differential equations originated from the study of surfaces in geometry and for solving a wide variety of problems in mechanics. During the second half of the nineteenth century, a large number of mathematicians became actively involved in the investigation of numerous problems presented by partial differential equations. The primary reason for this research was that partial differential equations both express many fundamental laws of nature and frequently arise in the mathematical analysis of diverse problems in science and engineering. The next phase of the development of linear partial differential equations is characterized by the efforts to develop the general theory and various methods of solutions of these linear equations. In fact, partial differential equations have been found to be essential to develop the theory of surfaces on the one hand and to the solution of physical problems on the other. These two areas of mathematics can be seen as linked by the bridge of the calculus of variations. With the discovery of the basic concepts and properties of distributions, the modern theory of the linear partial differential equations is now well established. The subject plays a central role in modern mathematics, especially in physics, geometry, and analysis. Although the origin of nonlinear partial differential equations is very old, they have undergone remarkable new developments during the last half of the twentieth century. One of the main impulses for developing nonlinear partial differential equations has been the study of nonlinear wave propagation problems. These problems arise in different areas of applied mathematics, physics, and engineering, including fluid dynamics, nonlinear optics, solid mechanics, plasma physics, quantum field theory, and condensed-matter physics. Nonlinear wave equations in particular have provided several examples of new solutions that are remarkably different from those obtained for linear wave problems. The best known examples of these are the corresponding shock waves, water waves, solitons and solitary waves. One of the remarkable properties of solitons is a localized wave form that is retained after interaction with other solitons, confirming solitons’ ‘particle-like’ behavior. Indeed, the theory of nonlinear waves and solitons has experienced a revolution over the past
xiv
Preface to the First Edition
three decades. During this revolution, many remarkable and unexpected phenomena have also been observed in physical, chemical, and biological systems. Other major achievements of twentieth-century applied mathematics include the discovery of soliton interactions, the Inverse Scattering Transform (IST) method for finding the explicit exact solution for several canonical partial differential equations, and asymptotic perturbation analysis for the investigation of nonlinear evolution equations. One of the major goals of the present book is to provide an accessible working knowledge of some of the current analytical methods required in modern mathematics, physics, and engineering. So the writing of the book was greatly influenced by the emphasis which Lord Rayleigh and Richard Feynman expressed as follows: In the mathematical investigation I have usually employed such methods as present themselves naturally to a physicist. The pure mathematician will complain, and (it must be confessed) sometimes with justice, of deficient rigor. But to this question there are two sides. For, however important it may be to maintain a uniformly high standard in pure mathematics, the physicist may occasionally do well to rest content with arguments which are fairly satisfactory and conclusive from his point of view. To his mind, exercised in a different order of ideas, the more severe procedure of the pure mathematician may appear not more but less demonstrative. And further, in many cases of difficulty to insist upon highest standard would mean the exclusion of the subject altogether in view of the space that would be required. Lord Rayleigh . . . However, the emphasis should be somewhat more on how to do the mathematics quickly and easily, and what formulas are true, rather than the mathematicians’ interest in methods of rigorous proof. Richard P. Feynman Audience and Organization This book provides an introduction to nonlinear partial differential equations and to the basic methods that have been found useful for finding the solutions of these equations. While teaching a course on partial differential equations, the author has had difficulty choosing textbooks to accompany the lectures on some modern topic in nonlinear partial differential equations. The book was developed as a result of many years of experience teaching partial differential equations at the senior undergraduate and/or graduate levels. Parts of this book have also been used to accompany lecturers on special topics in nonlinear partial differential equations at Indian universities during my recent visit on a Senior Fulbright Fellowship. Based on my experience, I believe that nonlinear partial differential equations are best approached through a sound knowledge of linear partial differential equations. In order to make the book self-contained, the first chapter deals with linear partial differential equations and their methods of solution with examples of applications. There is plenty of material in this book for a two-semester graduate level course for mathematics, science,
Preface to the First Edition
xv
and engineering students. Many new examples of applications to problems in fluid dynamics, plasma physics, nonlinear optics, gas dynamics, analytical dynamics, and acoustics are included. Special emphasis is given to physical, chemical, biological, and engineering problems involving nonlinear wave phenomena. It is not essential for the reader to have a thorough knowledge of the physical aspect of these topics, but limited knowledge of at least some of them would be helpful. Besides, the book is intended to serve as a reference work for those seriously interested in advanced study and research in the subject, whether for its own sake or for its applications to other fields of applied mathematics, mathematical physics, and engineering science. Another reason for adding this book to the literature is the fact that studies are continually being added to the theory, methods of solutions, and applications of nonlinear partial differential equations. It is becoming even more desirable for applied mathematicians, physicists, and engineering scientists to pursue study and research on these and related topics. Yet it is increasingly difficult to do so, because major articles appear in journals and research monographs of widely different natures. Some of these occur in papers scattered widely through the vast literature, and their connections are not readily apparent. This difficulty might be alleviated if a single book on nonlinear partial differential equations contained a coherent account of the recent developments, especially if written to be accessible to both graduate and postgraduate students. The field is growing fast. It is my hope that the book will first interest, then prepare readers to undertake research projects on nonlinear wave phenomena, reaction-diffusion phenomena, soliton dynamics, nonlinear instability and other nonlinear real-world problems, by providing that background of fundamental ideas, results, and methods essential to understanding the specialized literature of this vast area. The book is aimed at the reader interested in a broad perspective on the subject, the wide variety of phenomena encompassed by it and a working knowledge of the most important methods for solving the nonlinear equations. Those interested in more rigorous treatment of the mathematical theory of the subjects covered may consult some outstanding advanced books and treatises, listed in the Bibliography. Many ideas, principles, results, methods, examples of applications, and exercises presented in the book are either motivated by, or borrowed from works cited in the Bibliography. The author wishes to express his gratitude to the authors of these works. The first chapter provides an introduction to linear partial differential equations and to the methods of solutions of these equations, and to the basic properties of these solutions, that gives the reader a clear understanding of the subject and its varied examples of applications. Chapter 2 deals with nonlinear model equations and variational principles and the Euler–Lagrange equations. Included are variational principles for the nonlinear Klein–Gordon equation and for the nonlinear water waves. The third and fourth chapters are devoted to the first-order quasi-linear and nonlinear equations and to the method of characteristics for solving them. Examples of applications of these equations to analytical dynamics and nonlinear optics are included.
xvi
Preface to the First Edition
Chapters 5 and 6 deal with conservation laws and shock waves, and kinematic waves and specific real-world nonlinear problems. The concept of weak or discontinuous solutions is introduced in Section 5.4. Several sections of Chapter 6 discuss the properties of solutions of several real-world nonlinear models that include traffic flow, flood waves, chromatographic models, sediment transport in rivers, glacier flow, and roll waves. Chapter 7 is devoted to nonlinear dispersive waves, Whitham’s equations, and Whitham’s averaged variational principle. This is followed by the Whitham instability analysis and its applications to nonlinear water waves. In Chapter 8, we study the nonlinear diffusion-reaction phenomena, and Burgers’ and Fisher’s equations with physical applications. Special attention is given to traveling wave solutions and their stability analysis, similarity methods and similarity solutions of diffusion equations. Chapter 9 develops the theory of solitons and the Inverse Scattering Transform. Many recent results on the basic properties of the Korteweg–de Vries (KdV) and Boussinesq equations are discussed in some detail. Included are Bäcklund transformations, the nonlinear superposition principle, the Lax formulation and its KdV hierarchy. The nonlinear Schrödinger equation and solitary waves are the main focus of Chapter 10. Special attention is paid to examples of applications to fluid dynamics, plasma physics, and nonlinear optics. Chapter 11 is concerned with the theory of nonlinear Klein–Gordon and sineGordon equations with applications. The soliton and anti-soliton solutions of the sine-Gordon equation are described. The inverse scattering method, the similarity method and the method of separation of variables for the sine-Gordon equation are developed with examples. The final chapter deals with nonlinear evolution equations and asymptotic methods. Several asymptotic perturbation methods and the method of multiple scales are developed for the solutions of quasilinear dissipative systems, weakly and strongly dispersive systems. Salient Features The book contains 450 worked examples, examples of applications, and exercises which include some selected from many standard treatises as well as from recent research papers. It is hoped that they will serve as helpful self-tests for understanding of the theory and mastery of the nonlinear partial differential equations. These examples and examples of applications were chosen from the areas of partial differential equations, geometry, vibration and wave propagation, heat conduction in solids, electric circuits, dynamical systems, fluid mechanics, plasma physics, quantum mechanics, nonlinear optics, physical chemistry, mathematical modeling, population dynamics, and mathematical biology. This varied number of examples and exercises should provide something of interest for everyone. The exercises truly complement the text and range from the elementary to the challenging.
Acknowledgements
xvii
This book is designed as a new source for modern topics dealing with nonlinear phenomena and their applications for future development of this important and useful subject. Its main features are listed below: 1. A systematic mathematical treatment of some nonlinear partial differential equations, the methods of the solutions of these equations, and the basic properties of these solutions is presented, that gives the reader a clear understanding of the subject and its varied applications. 2. A detailed and clear explanation of every concept and method which is introduced, accompanied by carefully selected worked examples, is included with special emphasis being given to those topics in which students experience difficulty. 3. The book presents a wide variety of modern examples of applications carefully selected from areas of fluid dynamics, plasma physics, nonlinear optics, soliton dynamics, analytical dynamics, gas dynamics, and acoustics to provide motivation, and to illustrate the wide variety of real-world nonlinear problems. 4. Most of the recent developments in the subject since the early 1960s appear here in book form for the first time. 5. Included also is a broad coverage of the essential standard material on nonlinear partial differential equations and their applications that is not readily found in any texts or reference books. 6. A striking balance between the mathematical and physical aspects of the subject is maintained. 7. The book is organized with sufficient flexibility so as to enable instructors to select chapters according to length, emphasis and level of different courses. 8. A wide spectrum of exercises has been carefully chosen and included at the end of each chapter so the reader may further develop both manipulative skills in the applications of nonlinear equations and a deeper insight into this modern subject. 9. The book provides information that puts the reader at the forefront of current research. An updated Bibliography is included to stimulate new interest in future study and research. 10. Answers and hints to selected exercises with original source are provided at the end of the book for additional help to students.
Acknowledgements In preparing the book, the author has been encouraged by and has benefited from the helpful comments and criticism of a number of faculty, post-doctoral and doctoral students of several universities in the United States, Canada, and India. The author expresses his grateful thanks to these individuals for their interest in the book. My special thanks to Jackie Callahan and Ronee Trantham who typed the manuscript with many diagrams and cheerfully put up with constant changes and revisions. In spite of the best efforts of everyone involved, some typographical errors doubtless remain. I do hope that these are both few and obvious, and will cause minimum
xviii
Preface to the First Edition
confusion. Finally, the author wishes to express his special thanks to Mr. Wayne Yuhasz, Executive Editor, and the staff of Birkhäuser for their help and cooperation. I am deeply indebted to my wife, Sadhana, for her understanding and tolerance while the book was being written. University of Central Florida
Lokenath Debnath
Contents
1
Linear Partial Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Basic Concepts and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 The Linear Superposition Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.4 Some Important Classical Linear Model Equations . . . . . . . . . . . . . . . 8 1.5 Second-Order Linear Equations and Method of Characteristics . . . . 10 1.6 The Method of Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.7 Fourier Transforms and Initial Boundary-Value Problems . . . . . . . . . 34 1.8 Multiple Fourier Transforms and Partial Differential Equations . . . . 46 1.9 Laplace Transforms and Initial Boundary-Value Problems . . . . . . . . 51 1.10 Hankel Transforms and Initial Boundary-Value Problems . . . . . . . . . 61 1.11 Green’s Functions and Boundary-Value Problems . . . . . . . . . . . . . . . . 70 1.12 Sturm–Liouville Systems and Some General Results . . . . . . . . . . . . . 82 1.13 Energy Integrals and Higher Dimensional Equations . . . . . . . . . . . . . 101 1.14 Fractional Partial Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . 114 1.15 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2
Nonlinear Model Equations and Variational Principles . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Basic Concepts and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Some Nonlinear Model Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Variational Principles and the Euler–Lagrange Equations . . . . . . . . . 2.5 The Variational Principle for Nonlinear Klein–Gordon Equations . . 2.6 The Variational Principle for Nonlinear Water Waves . . . . . . . . . . . . . 2.7 The Euler Equation of Motion and Water Wave Problems . . . . . . . . . 2.8 The Energy Equation and Energy Flux . . . . . . . . . . . . . . . . . . . . . . . . . 2.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
First-Order, Quasi-linear Equations and Method of Characteristics . . 201 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 3.2 The Classification of First-Order Equations . . . . . . . . . . . . . . . . . . . . . 202
149 149 150 151 159 171 171 173 192 193
xx
Contents
3.3 3.4 3.5 3.6
The Construction of a First-Order Equation . . . . . . . . . . . . . . . . . . . . . The Geometrical Interpretation of a First-Order Equation . . . . . . . . . The Method of Characteristics and General Solutions . . . . . . . . . . . . Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
203 206 208 221
4
First-Order Nonlinear Equations and Their Applications . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 The Generalized Method of Characteristics . . . . . . . . . . . . . . . . . . . . . 4.3 Complete Integrals of Certain Special Nonlinear Equations . . . . . . . . 4.4 The Hamilton–Jacobi Equation and Its Applications . . . . . . . . . . . . . 4.5 Applications to Nonlinear Optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
227 227 228 231 238 246 254
5
Conservation Laws and Shock Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Conservation Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Discontinuous Solutions and Shock Waves . . . . . . . . . . . . . . . . . . . . . 5.4 Weak or Generalized Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
257 257 258 270 271 277
6
Kinematic Waves and Real-World Nonlinear Problems . . . . . . . . . . . . . 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Kinematic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Traffic Flow Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Flood Waves in Long Rivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 Chromatographic Models and Sediment Transport in Rivers . . . . . . . 6.6 Glacier Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.7 Roll Waves and Their Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . 6.8 Simple Waves and Riemann’s Invariants . . . . . . . . . . . . . . . . . . . . . . . 6.9 The Nonlinear Hyperbolic System and Riemann’s Invariants . . . . . . 6.10 Generalized Simple Waves and Generalized Riemann’s Invariants . . 6.11 The Lorenz System of Nonlinear Differential Equations and Deterministic Chaos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
283 283 284 286 297 300 305 307 311 329 338
Nonlinear Dispersive Waves and Whitham’s Equations . . . . . . . . . . . . . 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Linear Dispersive Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Initial-Value Problems and Asymptotic Solutions . . . . . . . . . . . . . . . . 7.4 Nonlinear Dispersive Waves and Whitham’s Equations . . . . . . . . . . . 7.5 Whitham’s Theory of Nonlinear Dispersive Waves . . . . . . . . . . . . . . . 7.6 Whitham’s Averaged Variational Principle . . . . . . . . . . . . . . . . . . . . . . 7.7 Whitham’s Instability Analysis of Water Waves . . . . . . . . . . . . . . . . . 7.8 Whitham’s Equation: Peaking and Breaking of Waves . . . . . . . . . . . . 7.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
355 355 356 359 361 364 367 369 371 377
7
342 346
Contents
xxi
8
Nonlinear Diffusion–Reaction Phenomena . . . . . . . . . . . . . . . . . . . . . . . . 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Burgers Equation and the Plane Wave Solution . . . . . . . . . . . . . . . . . . 8.3 Traveling Wave Solutions and Shock-Wave Structure . . . . . . . . . . . . . 8.4 The Exact Solution of the Burgers Equation . . . . . . . . . . . . . . . . . . . . 8.5 The Asymptotic Behavior of the Burgers Solution . . . . . . . . . . . . . . . 8.6 The N -Wave Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.7 Burgers Initial- and Boundary-Value Problem . . . . . . . . . . . . . . . . . . . 8.8 Fisher Equation and Diffusion–Reaction Process . . . . . . . . . . . . . . . . 8.9 Traveling Wave Solutions and Stability Analysis . . . . . . . . . . . . . . . . 8.10 Perturbation Solutions of the Fisher Equation . . . . . . . . . . . . . . . . . . . 8.11 Method of Similarity Solutions of Diffusion Equations . . . . . . . . . . . 8.12 Nonlinear Reaction–Diffusion Equations . . . . . . . . . . . . . . . . . . . . . . . 8.13 Brief Summary of Recent Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.14 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
381 381 382 384 386 391 392 394 396 399 402 404 412 416 417
9
Solitons and the Inverse Scattering Transform . . . . . . . . . . . . . . . . . . . . . 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 The History of the Solitons and Soliton Interactions . . . . . . . . . . . . . . 9.3 The Boussinesq and Korteweg–de Vries Equations . . . . . . . . . . . . . . . 9.4 Solutions of the KdV Equation: Solitons and Cnoidal Waves . . . . . . 9.5 The Lie Group Method and Similarity Analysis of the KdV Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.6 Conservation Laws and Nonlinear Transformations . . . . . . . . . . . . . . 9.7 The Inverse Scattering Transform (IST) Method . . . . . . . . . . . . . . . . . 9.8 Bäcklund Transformations and the Nonlinear Superposition Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.9 The Lax Formulation and the Zakharov and Shabat Scheme . . . . . . . 9.10 The AKNS Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.11 Asymptotic Behavior of the Solution of the KdV–Burgers Equation . 9.12 Strongly Dispersive Nonlinear Equations and Compactons . . . . . . . . 9.13 The Camassa–Holm (CH) and Degasperis–Procesi (DP) Nonlinear Model Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.14 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
425 425 426 431 458
10 The Nonlinear Schrödinger Equation and Solitary Waves . . . . . . . . . . . 10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2 The One-Dimensional Linear Schrödinger Equation . . . . . . . . . . . . . . 10.3 The Nonlinear Schrödinger Equation and Solitary Waves . . . . . . . . . 10.4 Properties of the Solutions of the Nonlinear Schrödinger Equation . . 10.5 Conservation Laws for the NLS Equation . . . . . . . . . . . . . . . . . . . . . . 10.6 The Inverse Scattering Method for the Nonlinear Schrödinger Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
535 535 536 537 541 547
466 469 474 494 499 506 508 509 518 528
550 552
xxii
Contents
10.8 Applications to Nonlinear Optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567 10.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575 11 Nonlinear Klein–Gordon and Sine-Gordon Equations . . . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 The One-Dimensional Linear Klein–Gordon Equation . . . . . . . . . . 11.3 The Two-Dimensional Linear Klein–Gordon Equation . . . . . . . . . . 11.4 The Three-Dimensional Linear Klein–Gordon Equation . . . . . . . . . 11.5 The Nonlinear Klein–Gordon Equation and Averaging Techniques 11.6 The Klein–Gordon Equation and the Whitham Averaged Variational Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.7 The Sine-Gordon Equation: Soliton and Antisoliton Solutions . . . . 11.8 The Solution of the Sine-Gordon Equation by Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.9 Bäcklund Transformations for the Sine-Gordon Equation . . . . . . . . 11.10 The Solution of the Sine-Gordon Equation by the Inverse Scattering Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.11 The Similarity Method for the Sine-Gordon Equation . . . . . . . . . . . 11.12 Nonlinear Optics and the Sine-Gordon Equation . . . . . . . . . . . . . . . 11.13 Nonlinear Lattices and the Toda-Lattice Soliton . . . . . . . . . . . . . . . . 11.14 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
579 579 580 582 584 585
12 Asymptotic Methods and Nonlinear Evolution Equations . . . . . . . . . . . 12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.2 The Reductive Perturbation Method and Quasi-linear Hyperbolic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3 Quasi-linear Dissipative Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.4 Weakly Nonlinear Dispersive Systems and the Korteweg–de Vries Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.5 Strongly Nonlinear Dispersive Systems and the NLS Equation . . . 12.6 The Perturbation Method of Ostrovsky and Pelinovsky . . . . . . . . . . 12.7 The Method of Multiple Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.8 Asymptotic Expansions and Method of Multiple Scales . . . . . . . . . 12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
623 623
13 Tables of Integral Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.1 Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2 Fourier Sine Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.3 Fourier Cosine Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.4 Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.5 Hankel Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.6 Finite Hankel Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
675 675 677 679 680 683 686
592 594 598 605 608 611 612 615 619
625 628 630 641 646 649 656 663 674
Contents
xxiii
A Some Special Functions and Their Properties . . . . . . . . . . . . . . . . . . . . . . A-1 Gamma, Beta, and Error Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . A-2 Bessel and Airy Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-3 Legendre and Associated Legendre Functions . . . . . . . . . . . . . . . . . A-4 Jacobi and Gegenbauer Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . A-5 Laguerre and Associated Laguerre Functions . . . . . . . . . . . . . . . . . . A-6 Hermite Polynomials and Weber–Hermite Functions . . . . . . . . . . . . A-7 Mittag-Leffler Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-8 The Jacobi Elliptic Integrals and Elliptic Functions . . . . . . . . . . . . .
689 689 697 703 706 710 712 713 715
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1 Fourier Series and Its Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . B-2 Generalized Functions (Distributions) . . . . . . . . . . . . . . . . . . . . . . . . B-3 Basic Properties of the Fourier Transforms . . . . . . . . . . . . . . . . . . . . B-4 Basic Properties of Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . .
719 719 740 751 758
C Answers and Hints to Selected Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . 1.15 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.14 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.14 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.14 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
767 767 785 788 794 796 799 803 805 807 808 809 809
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 813 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 847
1 Linear Partial Differential Equations
However varied may be the imagination of man, nature is still a thousand times richer, . . . . Each of the theories of physics . . . presents (partial differential) equations under a new aspect . . . without these theories, we should not know partial differential equations. Henri Poincaré Since a general solution must be judged impossible from want of analysis, we must be content with the knowledge of some special cases, and that all the more, since the development of various cases seems to be the only way to bringing us at last to a more perfect knowledge. Leonard Euler
1.1 Introduction Partial differential equations arise frequently in the formulation of fundamental laws of nature and in the mathematical analysis of a wide variety of problems in applied mathematics, mathematical physics, and engineering science. This subject plays a central role in modern mathematical sciences, especially in physics, geometry, and analysis. Many problems of physical interest are described by partial differential equations with appropriate initial and/or boundary conditions. These problems are usually formulated as initial-value problems, boundary-value problems, or initial boundary-value problems. In order to prepare the reader for study and research in nonlinear partial differential equations, a broad coverage of the essential standard material on linear partial differential equations and their applications is required. This chapter provides a review of basic concepts, principles, model equations, and their methods of solutions. This is followed by a systematic mathematical treatment of the theory and methods of solutions of second-order linear partial differential
L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_1,
2
1 Linear Partial Differential Equations
equations that gives the reader a clear understanding of the subject and its varied applications. Linear partial differential equations of the second order can be classified as one of the three types: hyperbolic, parabolic, and elliptic, and reduced to an appropriate canonical or normal form. The classification and method of reduction are described in Section 1.5. Special emphasis is given to various methods of solution of the initial-value and/or boundary-value problems associated with the three types of linear equations, each of which shows an entirely different behavior in properties and construction of solutions. Section 1.6 deals with the solutions of linear partial differential equations using the method of separation of variables combined with the superposition principle. A brief discussion of Fourier, Laplace, and Hankel transforms is included in Sections 1.7–1.10. These integral transforms are then applied to solve a large variety of initial and boundary problems described by partial differential equations. The transform solution combined with the convolution theorem provides an elegant representation of the solution for initial-value and boundary-value problems. Section 1.11 is devoted to Green’s functions for solving a wide variety of inhomogeneous partial differential equations of most common interest. This method can be made considerably easier by using generalized functions combined with appropriate integral transforms. The Sturm–Liouville systems and their general properties are discussed in Section 1.12. Section 1.13 deals with energy integrals, the law of conservation of energy, uniqueness theorems, and higher dimensional wave and diffusion equations. The final section contains some recent examples of fractional order diffusion–wave equations and their solutions.
1.2 Basic Concepts and Definitions A partial differential equation for a function u(x, y, . . .) is a relationship between u and its partial derivatives ux , uy , uxx , uxy , uyy , . . . , and can be written as F (x, y, u, ux , uy , uxx , uxy , uyy , . . .) = 0,
(1.2.1)
where F is some function, x, y, . . . are independent variables and u(x, y, . . .) is called a dependent variable. The order of a partial differential equation is defined in analogy with an ordinary differential equation as the highest-order of a derivative appearing in (1.2.1). The most general first-order partial differential equation can be written F (x, y, u, ux , uy ) = 0.
(1.2.2)
Similarly, the most general second-order partial differential equation in two independent variables x, y has the form F (x, y, u, ux , uy , uxx , uxy , uyy ) = 0,
(1.2.3)
and so on for higher-order equations. For example, x ux + y uy = 0,
(1.2.4)
1.2 Basic Concepts and Definitions
3
x ux + y uy = x2 + y 2 , u ux + ut = u, u2x + u2y = 1
(1.2.5) (1.2.6) (1.2.7)
uxx + 2uxy + uyy = 0, uxx + uyy = 0, utt − c2 uxx = f (x, t)
(1.2.8) (1.2.9)
are first-order equations, and
(1.2.10)
are second-order equations. Finally, ut + uux + uxxx = 0, utt + uxxxx = 0
(1.2.11) (1.2.12)
are examples of the third-order and fourth-order equations, respectively. A partial differential equation is called linear if it is linear in the unknown function and all its derivatives with coefficients depend only on the independent variables. It is called quasi-linear, if it is linear in the highest-order derivative of the unknown function. For example, (1.2.4), (1.2.5), (1.2.8)–(1.2.10) and (1.2.12) are linear equations, whereas (1.2.6) and (1.2.11) are quasi-linear equations. It is possible to write a partial differential equation in the operator form Lx u(x) = f (x),
(1.2.13)
where Lx is an operator. The operator Lx is called a linear operator if it satisfies the property Lx (au + bv) = a Lx u + b Lx v
(1.2.14)
for any two functions u and v and for any two constants a and b. Equation (1.2.13) is called linear if Lx is a linear operator. Equation (1.2.13) is called an inhomogeneous (or nonhomogeneous) linear equation. If f (x) ≡ 0, (1.2.13) is called a homogeneous equation. Equations (1.2.4), (1.2.8), (1.2.9), and (1.2.12) are linear homogeneous equations, whereas (1.2.5) and (1.2.10) are linear inhomogeneous equations. An equation which is not linear is called a nonlinear equation. If Lx is not linear, then (1.2.13) is called a nonlinear equation. Equations (1.2.6), (1.2.7), and (1.2.11) are examples of nonlinear equations. A classical solution (or simply a solution) of (1.2.1) is an ordinary function u = u(x, y, . . .) defined on some domain D which is continuously differentiable such that all its partial derivatives involved in the equation exist and satisfy (1.2.1) identically. However, this notion of classical solution can be extended by relaxing the requirement that u is continuously differentiable over D. The solution u = u(x, y, . . .) is called a weak (or generalized) solution of (1.2.1) if u or its partial derivatives are discontinuous at some or all points in D.
4
1 Linear Partial Differential Equations
To introduce the idea of a general solution of a partial differential equation, we solve a simple equation for u = u(x, y) of the form uxy = 0.
(1.2.15)
Integrating this equation with respect to x (keeping y fixed), we obtain uy = h(y), where h(y) is an arbitrary function of y. We then integrate it with respect to y to find u(x, y) = h(y) dy + f (x), where f (x) is an arbitrary function. Or equivalently, u(x, y) = f (x) + g(y),
(1.2.16)
where f (x) and g(y) are arbitrary functions. The solution (1.2.16) is called the general solution of the second-order equation (1.2.15). Usually, the general solution of a partial differential equation is an expression that involves arbitrary functions. This is a striking contrast to the general solution of an ordinary differential equation which involves arbitrary constants. Further, a simple equation (1.2.15) has infinitely many solutions. This can be illustrated by considering the problem of construction of partial differential equations from given arbitrary functions. For example, if u(x, t) = f (x − ct) + g(x + ct),
(1.2.17)
where f and g are arbitrary functions of (x − ct) and (x + ct), respectively, then uxx = f (x − ct) + g (x + ct), utt = c2 f (x − ct) + c2 g (x + ct) = c2 uxx , where primes denote differentiation with respect to the appropriate argument. Thus, we obtain the second-order linear equation, called the wave equation, utt − c2 uxx = 0.
(1.2.18)
Thus, the function u(x, t) defined by (1.2.17) satisfies (1.2.18) irrespective of the functional forms of f (x−ct) and g(x+ct), provided f and g are at least twice differentiable functions. Thus, the general solution of equation (1.2.18) is given by (1.2.17) which contains arbitrary functions. In the case of only two independent variables x, y, the solution u(x, y) of the equation (1.2.1) is visualized geometrically as a surface, called an integral surface in the (x, y, u) space.
1.3 The Linear Superposition Principle
5
1.3 The Linear Superposition Principle The general solution of a linear homogeneous ordinary differential equation of order n is a linear combination of n linearly independent solutions with n arbitrary constants. In other words, if u1 (x), u2 (x), . . . , un (x) are n linearly independent solutions of an nth order, linear, homogeneous, ordinary differential equation of the form Lu(x) = 0, (1.3.1) then, for any arbitrary constants c1 , c2 , . . . , cn , u(x) =
n
ck uk (x)
(1.3.2)
k=1
represents the most general solution of (1.3.1). This is called the linear superposition principle for ordinary differential equations. We note that the general solution of (1.3.1) depends on exactly n arbitrary constants. In the case of linear homogeneous partial differential equations of the form Lx u(x) = 0,
(1.3.3)
the general solution depends on arbitrary functions rather than arbitrary constants. So there are infinitely many solutions of (1.3.3). If we represent this infinite set of solutions of (1.3.3) by u1 (x), u2 (x), . . . , un (x), . . . , then the infinite linear combinations ∞ cn un (x), (1.3.4) u(x) = n=1
where cn are arbitrary constants, in general, may not be again a solution of (1.3.3) because the infinite series may not be convergent. So, for the case of partial differential equations, the superposition principle may not be true, in general. However, if there are only a finite number of solutions u1 (x), u2 (x), . . . , un (x) of the partial differential equation (1.3.3), then u(x) =
n
cn un (x)
(1.3.5)
n=1
again is a solution of (1.3.3) as can be verified by direct substitution. As with linear homogeneous ordinary differential equations, the principle of superposition applies to linear homogeneous partial differential equations and u(x) represents a solution of (1.3.3), provided that the infinite series (1.3.4) is convergent and the operator Lx can be applied to the series term by term. In order to generate such an infinite set of solutions un (x), the method of separation of variables is usually used. This method, combined with the superposition of solutions, is usually known as Fourier’s method, which will be described in a subsequent section.
6
1 Linear Partial Differential Equations
Another type of infinite linear combination is used to find the solution of a given partial differential equation. This is concerned with a family of solutions u(x, k) depending on a continuous real parameter k and a function c(k) such that
b
c(k)u(x, k) dk a
∞
or
c(k)u(x, k) dk
(1.3.6)
−∞
is convergent. Then, under certain conditions, this integral is again a solution. This may also be regarded as the linear superposition principle. In almost all cases, the general solution of a partial differential equation is of little use since it has to satisfy other supplementary conditions, usually called initial or boundary conditions. As indicated earlier, the general solution of a linear partial differential equation contains arbitrary functions. This means that there are infinitely many solutions and only by specifying the initial and/or boundary conditions can we determine a specific solution of interest. Usually, both initial and boundary conditions arise from the physics of the problem. In the case of partial differential equations in which one of the independent variables is the time t, an initial condition(s) specifies the physical state of the dependent variable u(x, t) at a particular time t = t0 or t = 0. Often u(x, 0) and/or ut (x, 0) are specified to determine the function u(x, t) at later times. Such conditions are called the Cauchy (or initial) conditions. It can be shown that these conditions are necessary and sufficient for the existence of a unique solution. The problem of finding the solution of the initial-value problem with prescribed Cauchy data on the line t = 0 is called the Cauchy problem or the initial-value problem. In each physical problem, the governing equation is to be solved within a given domain D of space with prescribed values of the dependent variable u(x, t) given on the boundary ∂D of D. Often, the boundary need not enclose a finite volume— in which case, part of the boundary is at infinity. For problems with a boundary at infinity, boundedness conditions on the behavior of the solution at infinity must be specified. This kind of problem is typically known as a boundary-value problem, and it is one of the most fundamental problems in applied mathematics and mathematical physics. There are three important types of boundary conditions which arise frequently in formulating physical problems. These are (a) Dirichlet conditions, where the solution u is prescribed at each point of a boundary ∂D of a domain D. The problem of finding the solution of a given equation Lx u(x) = 0 inside D with prescribed values of u on ∂D is called the Dirichlet boundary-value problem; ∂u of the solution on (b) Neumann conditions, where values of normal derivative ∂n the boundary ∂D are specified. In this case, the problem is called the Neumann boundary-value problem; ∂u + au) is specified on ∂D. The corresponding prob(c) Robin conditions, where ( ∂n lem is called the Robin boundary-value problem.
1.3 The Linear Superposition Principle
7
A problem described by a partial differential equation in a given domain with a set of initial and/or boundary conditions (or other supplementary conditions) is said to be well-posed (or properly posed) provided the following criteria are satisfied: (i) Existence: There exists at least one solution of the problem. (ii) Uniqueness: There is at most one solution. (iii) Stability: The solution must be stable in the sense that it depends continuously on the data. In other words, a small change in the given data must produce a small change in the solution. The stability criterion is essential for physical problems. A mathematical problem is usually considered physically realistic if a small change in given data produces correspondingly a small change in the solution. According to the Cauchy–Kowalewski theorem, the solution of an analytic Cauchy problem for partial differential equations exists and is unique. However, a Cauchy problem for Laplace’s equation is not always well-posed. A famous example of a non-well-posed (or ill-posed) problem was first given by Hadamard. Hadamard’s example deals with Cauchy’s initial-value problem for the Laplace equation ∇2 u ≡ uxx + uyy = 0,
0 < y < ∞, x ∈ R
(1.3.7)
1 u(x, 0) = 0 and uy (x, 0) = sin nx, n
(1.3.8)
with the Cauchy data
where n is an integer representing the wavenumber. These data tend to zero uniformly as n → ∞. It can easily be verified that the unique solution of this problem is given by 1 sinh ny sin nx. (1.3.9) u(x, y) = n2 As n → ∞, this solution does not tend to the solution u = 0. In fact, solution (1.3.9) represents oscillations in x with unbounded amplitude n−2 sinh ny which tends to infinity as n → ∞. In other words, although the data change by an arbitrarily small amount, the change in the solution is infinitely large. So the problem is certainly not well-posed, that is, the solution does not depend continuously on the initial data. Even if the wavenumber n is a fixed, finite quantity, the solution is clearly unstable in the sense that u(x, y) → ∞ as y → ∞ for any fixed x, such that sinc nx = 0. On the other hand, the Cauchy problem (see Example 1.5.3) for the simplest hyperbolic equation (1.5.29) with the initial data (1.5.35ab) is a well posed problem. As to the domain of dependence for the solution, u(x, t) depends only on those values of f (ξ) and g(ξ) for which x − ct ≤ ξ ≤ x + ct. Similarly, the Cauchy problems for parabolic equations are generally well posed. We conclude this section with a general remark. The existence, uniqueness, and stability of solutions are the basic requirements for a complete description of a physical problem with appropriate initial and boundary conditions. However, there are
8
1 Linear Partial Differential Equations
many situations in applied mathematics which deal with ill-posed problems. In recent years, considerable progress has been made on the theory of ill-posed problems, but the discussion of such problems is beyond the scope of this book.
1.4 Some Important Classical Linear Model Equations We start with a special type of second-order linear partial differential equation for the following reasons. First, second-order equations arise more frequently in a wide variety of applications. Second, their mathematical treatment is simpler and easier to understand than that of first-order equations, in general. Usually, in almost all physical phenomena, the dependent variable u = u(x, y, z, t) is a function of the three space variables and the time variable t. Included here are only examples of equations of most common interest. Example 1.4.1. The wave equation is utt − c2 ∇2 u = 0,
(1.4.1)
where
∂2 ∂2 ∂2 + 2 + 2, (1.4.2) 2 ∂x ∂y ∂z and c is a constant. This equation describes the propagation of a wave (or disturbance), and it arises in a wide variety of physical problems. Some of these problems include a vibrating string, vibrating membrane, longitudinal vibrations of an elastic rod or beam, shallow water waves, acoustic problems for the velocity potential for a fluid flow through which sound can be transmitted, transmission of electric signals along a cable, and both electric and magnetic fields in the absence of charge and dielectric. ∇2 ≡
Example 1.4.2. The heat or diffusion equation is ut − κ∇2 u = 0,
(1.4.3)
where κ is the constant of diffusivity. This equation describes the diffusion of thermal energy in a homogeneous medium. It can be used to model the flow of a quantity, such as heat, or a concentration of particles. It is also used as a model equation for growth and diffusion, in general, and growth of a solid tumor, in particular. The diffusion equation describes the unsteady boundary-layer flow in the Stokes and Rayleigh problems and also the diffusion of vorticity from a vortex sheet. Example 1.4.3. The Laplace equation is ∇2 u = 0.
(1.4.4)
This equation is used to describe electrostatic potential in the absence of charges, gravitational potential in the absence of mass, equilibrium displacement of an elastic membrane, velocity potential for an incompressible fluid flow, temperature in a steady-state heat conduction problem, and many other physical phenomena.
1.4 Some Important Classical Linear Model Equations
9
Example 1.4.4. The Poisson equation is ∇2 u = f (x, y, z),
(1.4.5)
where f (x, y, z) is a given function describing a source or sink. This is an inhomogeneous Laplace equation, and hence, the Poisson equation is used to study all phenomena described by the Laplace equation in the presence of external sources or sinks. Example 1.4.5. The Helmholtz equation is ∇2 u + λu = 0,
(1.4.6)
where λ is a constant. This is a time-independent wave equation (1.4.1) with λ as a separation constant. In particular, its solution in acoustics represents an acoustic radiation potential. Example 1.4.6. The telegraph equation is given in a general form as utt − c2 uxx + aut + bu = 0,
(1.4.7)
where a, b, and c are constants. This equation arises in the study of propagation of electrical signals in a cable of a transmission line. Both current I and voltage V satisfy an equation of the form (1.4.7). This equation also arises in the propagation of pressure waves in the study of pulsatile blood flow in arteries and in one-dimensional random motion of bugs along a hedge. Example 1.4.7. The Klein–Gordon (or KG) equation is 2 2 mc ψ + ψ = 0,
(1.4.8)
where
∂2 − c 2 ∇2 (1.4.9) ∂t2 is the d’Alembertian operator, h(= 2π) is the Planck constant, and m is a constant mass of the particle. Klein (1927) and Gordon (1926) derived a relativistic equation for a charged particle in an electromagnetic field. It is of conservative dispersive type and played an important role in our understanding of the elementary particles. This equation is also used to describe dispersive wave phenomena, in general. ≡
Example 1.4.8. The time-independent Schrödinger equation in quantum mechanics is 2 ∇2 ψ + (E − V )ψ = 0, (1.4.10) 2m where h(= 2π) is the Planck constant, m is the mass of the particle whose wave function is ψ(x, y, z, t), E is a constant, and V is the potential energy. If V = 0, (1.4.10) reduces to the Helmholtz equation.
10
1 Linear Partial Differential Equations
Example 1.4.9. The linear Korteweg–de Vries (or KdV) equation is ut + αux + βuxxx = 0,
(1.4.11)
where α and β are constants. This describes the propagation of linear, long, water waves and of plasma waves in a dispersive medium. Example 1.4.10. The linear Boussinesq equation is utt − α2 ∇2 u − β 2 ∇2 utt = 0,
(1.4.12)
where α and β are constants. This equation arises in elasticity for longitudinal waves in bars, long water waves, and plasma waves. Example 1.4.11. The biharmonic wave equation is utt + c2 ∇4 u = 0,
(1.4.13)
where c is a constant. In elasticity, the displacement of a thin elastic plate by small vibrations satisfies this equation. When u is independent of time t, (1.4.13) reduces to what is called the biharmonic equation, namely ∇4 u = 0.
(1.4.14)
This describes the equilibrium equation for the distribution of stresses in an elastic medium satisfied by Airy’s stress function u(x, y, z). In fluid dynamics, this equation is satisfied by the stream function ψ(x, y, z) in a viscous fluid flow. Example 1.4.12. The electromagnetic wave equations for the electric field E and the polarization P are (1.4.15) E0 Ett − c20 Exx + Ptt = 0, 2 2 Ptt + ω0 P − E0 ωp E = 0, (1.4.16) where E0 is the permittivity (or dielectric constant) of free space, ω0 is the natural frequency of the oscillator, c0 is the speed of light in a vacuum, and ωp is the plasma frequency.
1.5 Second-Order Linear Equations and Method of Characteristics The general second-order linear partial differential equation in two independent variables x, y is given by Auxx + Buxy + Cuyy + Dux + Euy + F u = G,
(1.5.1)
1.5 Second-Order Linear Equations and Method of Characteristics
11
where A, B, C, D, E, F , and G are given functions of x and y or constants. The classification of second-order equations is based upon the possibility of reducing equation (1.5.1) by a coordinate transformation to a canonical or standard form at a point. We consider the transformation from x, y to ξ, η defined by ξ = φ(x, y),
η = ψ(x, y),
(1.5.2ab)
where φ and ψ are twice continuously differentiable and the Jacobian J(x, y) = φx ψy − ψx φy is nonzero in a domain of interest so that x, y can be determined uniquely from the system (1.5.2ab). Then, by the chain rule, ux = uξ ξx + uη ηx , uy = uξ ξy + uη ηy , (ux )x = (ux )ξ ξx + (ux )η ηx = (uξ ξx + uη ηx )ξ ξx + (uξ ξx + uη ηx )η ηx , = uξξ ξx2 + 2uξη ξx ηx + uηη ηx2 + uξ ξxx + uη ηxx ,
uxx uyy = uξξ ξy2 + 2uξη ξy ηy + uηη ηy2 + uξ ξyy + uη ηyy , uxy = uξξ ξx ξy + uξη (ξx ηy + ξy ηx ) + uηη ηx ηy + uξ ξxy + uη ηxy . Substituting these results into equation (1.5.1) gives A∗ uξξ + B ∗ uξη + C ∗ uηη + D∗ uξ + E ∗ uη + F ∗ u = G∗ ,
(1.5.3)
where A∗ = Aξx2 + Bξx ξy + Cξy2 , B ∗ = 2Aξx ηx + B(ξx ηy + ξy ηx ) + 2Cξy ηy , C ∗ = Aηx2 + Bηx ηy + Cηy2 , D ∗ = Aξxx + Bξxy + Cξyy + Dξx + Eξy , E ∗ = Aηxx + Bηxy + Cηyy + Dηx + Eηy , F ∗ = F, and G∗ = G. Now, the problem is to determine ξ and η so that equation (1.5.3) takes the simplest possible form. We choose ξ and η such that A∗ = C ∗ = 0 and B ∗ = 0. Or, more explicitly, A∗ = Aξx2 + Bξx ξy + Cξy2 = 0, ∗
C =
Aηx2
+ Bηx ηy +
Cηy2
= 0.
(1.5.4) (1.5.5)
These two equations can be combined into a single quadratic equation for ζ = ξ or η 2 ζx ζx + C = 0. (1.5.6) A +B ζy ζy We consider the level curves ξ = φ(x, y) = const. = C1 and η = ψ(x, y) = const. = C2 . On these curves
12
1 Linear Partial Differential Equations
dξ = ξx dx + ξy dy = 0,
dη = ηx dx + ηy dy = 0,
(1.5.7ab)
that is, the slopes of these curves are given by ξx dy =− , dx ξy
ηx dy =− . dx ηy
(1.5.8ab)
Thus, the slopes of both level curves are the roots of the same quadratic equation which is obtained from (1.5.6) as 2 dy dy −B + C = 0, (1.5.9) A dx dx and the roots of this equation are given by 1 dy = B ± B 2 − 4AC . (1.5.10ab) dx 2A These equations are known as the characteristic equations for (1.5.1), and their solutions are called the characteristic curves, or simply the characteristics of equation (1.5.1). The solution of the two ordinary differential equations (1.5.10ab) defines two distinct families of characteristics φ(x, y) = C1 and ψ(x, y) = C2 . There are three possible cases to consider. Case I. B 2 − 4AC > 0. Equations for which B 2 − 4AC > 0 are called hyperbolic. Integrating (1.5.10ab) gives two real and distinct families of characteristics φ(x, y) = C1 and ψ(x, y) = C2 , where C1 and C2 are constants of integration. Since A∗ = C ∗ = 0, and B ∗ = 0, and dividing by B ∗ , equation (1.5.3) reduces to the form 1 uξη = − ∗ D ∗ uξ + E ∗ uη + F ∗ u − G∗ = H1 (say). (1.5.11) B This is called the first canonical form of the hyperbolic equation. If the new independent variables α = ξ + η, β =ξ−η (1.5.12ab) are introduced, then uη = uα αη + uβ βη = uα − uβ , uξ = uα αξ + uβ βξ = uα + uβ , (uη )ξ = (uη )α αξ + (uη )β βξ = (uα − uβ )α · 1 + (uα − uβ )β · 1 = uαα − uββ . Consequently, equation (1.5.11) becomes uαα − uββ = H2 (α, β, u, uα , uβ ).
(1.5.13)
This is called the second canonical form of the hyperbolic equation. It is important to point out that characteristics play a fundamental role in the theory of hyperbolic equations. Case II. B 2 − 4AC = 0. There is only one family of real characteristics whose slope, due to (1.5.10ab), is given by
1.5 Second-Order Linear Equations and Method of Characteristics
13
dy B = . (1.5.14) dx 2A Integrating this equation gives ξ = φ(x, y) = const. (or η = ψ(x, y) = const.). Since B 2 = 4AC and A∗ = 0, we obtain √ 2 √ 0 = A∗ = Aξx2 + Bξx ξy + Cξy2 = Aξx + Cξy . It then follows that B ∗ = 2Aξx ηx + B(ξx ηy + ξy ηx ) + 2Cξy ηy √ √ √ √ = 2 Aξx + Cξy Aηx + Cηy = 0 for an arbitrary value of η which is independent of ξ. For example, if η = y, the Jacobian is nonzero in the domain of parabolicity. Dividing (1.5.3) by C ∗ = 0 yields uηη = H3 (ξ, η, u, uξ , uη ).
(1.5.15)
This is known as the canonical form of the parabolic equation. On the other hand, if we choose η = ψ(x, y) = const. as the integral of (1.5.14), equation (1.5.3) assumes the form uξξ = H3∗ (ξ, η, u, uξ , uη ).
(1.5.16)
Equations for which B 2 − 4AC = 0 are called parabolic. Case III. B 2 − 4AC < 0. Equations for which B 2 − 4AC < 0 are called elliptic. In this case, equations (1.5.10ab) have no real solutions. So there are two families of complex characteristics. Since the roots ξ, η of (1.5.10ab) are complex conjugates of each other, we introduce the new real variables as α=
1 (ξ + η), 2
β=
1 (ξ − η), 2i
(1.5.17ab)
so that ξ = α + iβ and η = α − iβ. We use (1.5.17ab) to transform (1.5.3) into the form A∗∗ uαα + B ∗∗ uαβ + C ∗∗ uββ = H4 (α, β, u, uα , uβ ),
(1.5.18)
where the coefficients of this equation assume the same form as the coefficients of (1.5.3). It can easily be verified that A∗ = 0 and C ∗ = 0 take the form A∗∗ − C ∗∗ ± iB ∗∗ = 0, which are satisfied if and only if A∗∗ = C ∗∗
and
B ∗∗ = 0.
14
1 Linear Partial Differential Equations
Thus, dividing by A∗∗ , equation (1.5.18) reduces to the form uαα + uββ =
H4 = H5 (α, β, u, uα , uβ ). A∗∗
(1.5.19)
This is called the canonical form of the elliptic equation. In summary, we state that equation (1.5.1) is called hyperbolic, parabolic, or elliptic at a point (x0 , y0 ) accordingly as B 2 (x0 , y0 ) − 4A(x0 , y0 )C(x0 , y0 ) > = < 0.
(1.5.20)
If it is true at all points in a given domain, then the equation is said to be hyperbolic, parabolic, or elliptic in that domain. Finally, it has been shown above that, for the case of two independent variables, a transformation can always be found to transform the given equation to the canonical form. However, in the case of several independent variables, in general, it is not possible to find such a transformation. These three types of partial differential equations arise in many areas of mathematical and physical sciences. Usually, boundary-value problems are associated with elliptic equations, whereas the initial-value problems arise in connection with hyperbolic and parabolic equations. Example 1.5.1. Show that (a) the wave equation utt − c2 uxx = 0 is hyperbolic, (b) the diffusion equation ut − κuxx = 0 is parabolic, (c) the Laplace equation uxx + uyy = 0 is elliptic, (d) the Tricomi equation uxx + xuyy = 0 is elliptic for x > 0, parabolic for x = 0, and hyperbolic for x < 0. For case (a), A = −c2 , B = 0, and C = 1. Hence, B 2 − 4AC = c2 > 0 for all x and t. So, the wave equation is hyperbolic everywhere. Similarly, the reader can show (b) and (c). Finally, for case (d), A = 1, B = 0, C = x, hence, B 2 − 4AC = −4x < 0, = 0, or > 0 accordingly as x > 0, x = 0, or x < 0, and the result follows. Example 1.5.2. Find the characteristic equations and characteristics and then reduce the equation xuxx + uyy = x2 (1.5.21) to canonical form. In this problem, A = x, B = 0, C = 1, B 2 − 4AC = −4x. Thus, the equation is hyperbolic if x < 0, parabolic if x = 0, and elliptic if x > 0. The characteristics equations are √ B ± B 2 − 4AC 1 dy = = ±√ . (1.5.22ab) dx 2A −x Hence,
√ √ y = ±2 −x = const. = ±2 −x + c,
1.5 Second-Order Linear Equations and Method of Characteristics
15
Fig. 1.1 Characteristics are parabolas for x < 0.
or equivalently, √ ξ = y + 2 −x = const.,
√ η = y − 2 −x = const.
(1.5.23ab)
These represent two branches of the parabolas (y − c)2 = −4x where c is a constant. The former equation (ξ = const.) gives a branch with positive slopes, whereas the latter equation (η = const.) represents a branch with negative slopes as shown in Figure 1.1. Both branches are tangent to the y-axis which is the single characteristic in the parabolic region. Indeed, the y-axis is the envelope of the characteristics for the hyperbolic region x < 0. For x < 0, we use the transformations √ ξ = y + 2 −x,
√ η = y − 2 −x
(1.5.24ab)
to reduce (1.5.21) to the canonical form. We find ξx = − √
1 , −x
ξy = 1,
ξxx = −
ηx = + √
1 , −x
ηy = 1,
ηxx =
√ (ξ − η) = 4 −x,
and
xuxx + uyy = x2 reduces to the form
1 1 , 2 (−x)3/2
(ξ − η)4 = (16x)2 .
Consequently, the equation
1 1 , 2 (−x)3/2
ξyy = 0, ηyy = 0,
16
1 Linear Partial Differential Equations
x uξξ ξx2 + 2uξη ξx ηx + uηη ηx2 + uξ ξxx + uη ηxx + uξξ ξy2 + 2uξη ξy ηy + uηη ηy2 + uξ ξyy + uη ηyy = x2 . Or equivalently,
1 1 1 1 1 1 1 x uξξ − + 2uξη − uηη − uξ + uη x x x 2 (−x)3/2 2 (−x)3/2 + [uξξ + 2uξη + uηη ] = x2 . This leads to the transformed equation which is 4uξη +
1 1 1 √ (uξ − uη ) = (ξ − η)4 , 2 −x (16)2
or equivalently, uξη =
1 1 1 1 · (uξ − uη ). (ξ − η)4 − 4 (16)2 2 (ξ − η)
This is the first canonical form. For x > 0, we use the transformations √ ξ = y + 2i x, so that α=
1 (ξ + η) = y, 2
β=
(1.5.25)
√ η = y − 2i x √ 1 (ξ − η) = 2 x. 2i
(1.5.26ab)
Clearly, αx = 0, 1 βx = √ , x
αy = 1,
αxx = 0,
βy = 0,
βxx
αyy = 0, αxy = 0, 1 1 = − 3/2 , βyy = 0. 2x
So, equation (1.5.21) reduces to the canonical form x uαα αx2 + 2uαβ αx βx + uββ βx2 + uα αxx + uβ βxx +
uαα αy2
+ 2uαβ αy βy +
uββ βy2
+ uα αyy + uβ βyy
or uαα + uββ −
1 1 √ uβ = 2 x
uαα + uββ
4 β 2
1 = uβ + β
4 β . 2
4 β = , 2
(1.5.27)
This is the desired canonical form of the elliptic equation. Finally, for the parabolic case (x = 0), equation (1.5.21) reduces to the canonical form
1.5 Second-Order Linear Equations and Method of Characteristics
17
(1.5.28)
uyy = 0. dx dy
In this case, the characteristic determined from = 0 is x = 0. That is, the y-axis is the characteristic curve, and it represents a curve across which a transition from hyperbolic to elliptic form takes place. Example 1.5.3 (The Cauchy Problem for the Wave Equation). The one-dimensional wave equation (1.5.29) utt − c2 uxx = 0 is a special case of (1.5.1) with A = −c2 , B = 0, and C = 1. Hence, B 2 − 4AC = 4c2 > 0, and therefore, the equation is hyperbolic, as mentioned before. According to (1.5.10ab), the equations of characteristics are dt 1 =± , dx c
(1.5.30)
or equivalently, ξ = x − ct = const.,
η = x + ct = const.
(1.5.31ab)
This shows that the characteristics are straight lines in the (x, t)-plane. The former represents a family of lines with positive slopes, and the latter a family of lines with negative slopes in the (x, t)-plane. In terms of new coordinates ξ and η, we obtain uxx = uξξ + 2uξη + uηη ,
utt = c2 (uξξ − 2uξη + uηη )
so that the wave equation (1.5.29) becomes −4c2 uξη = 0.
(1.5.32)
Since c = 0, uξη = 0 which can be integrated twice to obtain the solution u(ξ, η) = φ(ξ) + ψ(η),
(1.5.33)
where φ and ψ are arbitrary functions. Thus, in terms of the original variables, we obtain u(x, t) = φ(x − ct) + ψ(x + ct). (1.5.34) This represents the general solution provided φ and ψ are arbitrary but twice differentiable functions. The first term φ(x − ct) represents a wave (or disturbance) traveling to the right with constant speed c. Similarly, ψ(x + ct) represents a wave moving to the left with constant speed c. Thus, the general solution u(x, t) is a linear superposition of two such waves. The typical initial-value problem for the wave equation (1.5.29) is the Cauchy problem of an infinite string with initial data u(x, 0) = f (x),
ut (x, 0) = g(x),
(1.5.35ab)
where f (x) and g(x) are given functions representing the initial displacement and initial velocity, respectively. The conditions (1.5.35ab) imply that
18
1 Linear Partial Differential Equations
φ(x) + ψ(x) = f (x), −cφ (x) + cψ (x) = g(x),
(1.5.36) (1.5.37)
where the prime denotes the derivative with respect to the argument. Integrating equation (1.5.37) gives x g(τ ) dτ, (1.5.38) −cφ(x) + cψ(x) = x0
where x0 is an arbitrary constant. Equations (1.5.36) and (1.5.38) now yield 1 1 x g(τ ) dτ, φ(x) = f (x) − 2 2c x0 1 x 1 ψ(x) = f (x) + g(τ ) dτ. 2 2c x0 Obviously, (1.5.34) gives the so-called d’Alembert solution of the Cauchy problem as 1 1 x+ct g(τ ) dτ. (1.5.39) u(x, t) = f (x − ct) + f (x + ct) + 2 2c x−ct It can be verified by direct substitution that u(x, t) satisfies equation (1.5.29) provided f is twice differentiable and g is differentiable. Further, the d’Alembert solution (1.5.39) can be used to show that this problem is well posed. The solution (1.5.39) consists of terms involving f (x ± ct) and the term involving the integral of g. Both terms combined together suggest that the value of the solution at position x and time t depends only on the initial values of f (x) at points x ± ct and the value of the integral of g between these points. The interval (x − ct, x + ct) is called the domain of dependence of (x, t). The terms involving f (x ± ct) in (1.5.39) show that waves are propagated along the characteristics with constant velocity c. In particular, if g(x) = 0, the solution is represented by the first two terms in (1.5.39), that is, 1 (1.5.40) u(x, t) = f (x − ct) + f (x + ct) . 2 Physically, this solution shows that the initial data is split into two equal waves, similar in shape to the initial displacement, but of half the amplitude. These waves propagate in the opposite direction with the same constant speed c as shown in Figure 1.2. To investigate the physical significance of the d’Alembert solution, it is convenient to rewrite the solution in the form 1 x−ct 1 1 f (x − ct) − g(τ ) dτ + f (x + ct) u(x, t) = 2 2c 0 2 x+ct 1 + g(τ ) dτ (1.5.41) 2c 0 = Φ(x − ct) + Ψ (x + ct), (1.5.42)
1.5 Second-Order Linear Equations and Method of Characteristics
19
Fig. 1.2 Splitting of initial data into equal waves.
Fig. 1.3 Graphical representation of solution
where
ξ 1 1 Φ(ξ) = f (ξ) − g(τ ) dτ, 2 2c 0 η 1 1 Ψ (η) = f (η) + g(τ ) dτ. 2 2c 0
(1.5.43a) (1.5.43b)
Physically, Φ(x − ct) represents a progressive wave propagating in the positive xdirection with constant speed c without change of shape as shown in Figure 1.3. Similarly, Ψ (x + ct) also represents a progressive wave traveling in the negative x-direction with the same speed without change of shape. A more general form of the wave equation is utt − a2 (x)uxx = 0,
(1.5.44)
where a is a function of x only. The characteristic coordinates are now given by
20
1 Linear Partial Differential Equations
ξ =t−
x
dτ , a(τ )
x
η =t+
dτ . a(τ )
(1.5.45ab)
Thus, 1 1 u x = − uξ + u η , ut = uξ + uη , a a 1 a (x) uxx = 2 (uξξ − 2uξη + uηη ) − (uη − uξ ) 2 , a a utt = uξξ + 2uξη + uηη . Consequently, equation (1.5.44) reduces to 4uξη + a (x)(uη − uξ ) = 0.
(1.5.46)
In order to express a in terms of ξ and η, we observe that x dτ , η−ξ =2 a(τ )
(1.5.47)
so that x is a function of (η − ξ). Thus, a (x) will be some function of (η − ξ). In particular, if a(x) = Axn , where A is a constant, so that a (x) = nAxn−1 , and when n = 1, result (1.5.47) gives η−ξ =−
1 1 2 n−1 A (n − 1) x
a (x) = −
2n 1 · . (n − 1) η − ξ
so that
(1.5.48)
Thus, equation (1.5.46) reduces to the form 4uξη −
2n 1 (uη − uξ ) = 0. (n − 1) (η − ξ)
Finally, we find that uξη =
n 1 (uη − uξ ). 2(n − 1) (η − ξ)
When n = 1, a(x) = Ax, and a (x) = A, substituting ξ = used to reduce equation (1.5.46) to uαβ =
1 (uα − uβ ). 4
(1.5.49) α A
and η =
β A
can be
(1.5.50)
Equation (1.5.49) is called the Euler–Darboux equation which has the hyperbolic form m uxy = (ux − uy ), (1.5.51) x−y where m is a positive integer.
1.6 The Method of Separation of Variables
21
We next note that
∂u ∂ ∂2 (x − y) (x − y)u = − u = (x − y)uxy + (uy − ux ). (1.5.52) ∂x∂y ∂x ∂y When m = 1, equation (1.5.51) becomes (x − y)uxy = ux − uy so that (1.5.52) reduces to ∂2 (x − y)u = 0. ∂x∂y
(1.5.53)
This shows that the solution of (1.5.53) is (x − y)u = φ(x) + ψ(y). Hence, the solution of (1.5.51) with m = 1 is u(x, y) =
φ(x) + ψ(y) , x−y
(1.5.54)
where φ and ψ are arbitrary functions. ∂2 , so that the result We multiply (1.5.51) by (x − y), and apply the derivative ∂x∂y is, due to (1.5.52), ∂ ∂ ∂2 ∂ ∂ (uxy ) = m uxy . (uxy ) + − − (x − y) ∂x∂y ∂y ∂x ∂x ∂y Or
∂ ∂2 ∂ (x − y) (uxy ). (uxy ) = (m + 1) − ∂x∂y ∂x ∂y
(1.5.55)
Hence, if u is a solution of (1.5.51), then uxy is a solution of (1.5.51) with m replaced by m + 1. When m = 1, the solution is given by (1.5.54), and hence, the solution of (1.5.51) takes the form
∂ 2m−2 φ(x) + ψ(y) u(x, y) = , (1.5.56) ∂xm−1 ∂y m−1 x−y where φ and ψ are arbitrary functions.
1.6 The Method of Separation of Variables This method combined with the principle of superposition is widely used to solve initial boundary-value problems involving linear partial differential equations. Usually, the dependent variable u(x, y) is expressed in the separable form as u(x, y) = X(x)Y (y) where X and Y are functions of x and y, respectively. In many cases, the partial differential equation reduces to two ordinary differential equations for X and Y . A similar treatment can be applied to equations in three or more independent
22
1 Linear Partial Differential Equations
variables. However, the question of separability of a partial differential equation into two or more ordinary differential equations is by no means a trivial one. In spite of this question, the method is widely used in finding solutions of a large class of initial boundary-value problems. This method of solution is known as the Fourier method (or the method of eigenfunction expansion). Thus, the procedure outlined above leads to the important ideas of eigenvalues, eigenfunctions, and orthogonality, all of which are very general and powerful for dealing with linear problems. The following examples illustrate the general nature of this method. Example 1.6.1 (Transverse Vibration of a String). We consider the one-dimensional linear wave equation utt = c2 uxx ,
0 < x < , t > 0,
(1.6.1)
where c2 = T ∗ /ρ, T ∗ is a constant tension, and ρ is the constant line density of the string. The initial and boundary conditions are u(x, 0) = f (x), ut (x, 0) = g(x), u(0, t) = 0 = u(, t), t > 0,
0 ≤ x ≤ ,
(1.6.2ab) (1.6.3ab)
where f and g are the initial displacement and initial velocity, respectively. According to the method of separation of variables, we assume a separable solution of the form u(x, t) = X(x)T (t) = 0, (1.6.4) where X is a function of x only, and T is a function of t only. Substituting this solution into equation (1.6.1) yields 1 d2 T 1 d2 X = 2 . 2 X dx c T dt2
(1.6.5)
Since the left side of this equation is a function of x only and the right-hand side is a function of t only, it follows that (1.6.5) can be true only if both sides are equal to the same constant value. We then write 1 d2 T 1 d2 X = = λ, X dx2 c2 T dt2
(1.6.6)
where λ is an arbitrary separation constant. Thus, this leads to the pair of ordinary differential equations d2 X = λX, dx2 d2 T = λc2 T. dt2
(1.6.7a) (1.6.7b)
We solve this pair of equations by using the boundary conditions which are obtained from (1.6.3ab) as
1.6 The Method of Separation of Variables
u(0, t) = X(0)T (t) = 0 for t > 0, u(, t) = X()T (t) = 0 for t > 0.
23
(1.6.8) (1.6.9)
Hence, we take T (t) = 0 to obtain X(0) = 0 = X().
(1.6.10ab)
There are three possible cases: (i) λ > 0, (ii) λ = 0, (iii) λ < 0. For case (i), λ = α2 > 0. The solution of (1.6.7a) is X(x) = Aeαx + Be−αx ,
(1.6.11)
which together with (1.6.10ab) leads to A = B = 0. This leads to a trivial solution u(x, t) = 0. For case (ii), λ = 0. In this case, the solution of (1.6.7a) is X(x) = Ax + B.
(1.6.12)
Then. we use (1.6.10ab) to obtain A = B = 0. This also leads to the trivial solution u(x, t) = 0. For case (iii), λ < 0, and hence, we write λ = −α2 so that the solution of equation (1.6.7a) gives X = A cos αx + B sin αx, (1.6.13) whence, using (1.6.10ab), we derive the nontrivial solution X(x) = B sin αx,
(1.6.14)
where B is an arbitrary nonzero constant. Clearly, since B = 0 and X() = 0, sin α = 0, which gives the solution for the parameter α nπ , n = 1, 2, 3, . . . . α = αn =
(1.6.15)
(1.6.16)
Note that n = 0 (α = 0) leads to the trivial solution u(x, t) = 0, and hence, the case n = 0 is to be excluded. Evidently, we see from (1.6.16) that there exists an infinite set of discrete values of α for which the problem has a nontrivial solution. These values αn are called the eigenvalues, and the corresponding solutions are nπx . (1.6.17) Xn (x) = Bn sin We next solve (1.6.7a) with λ = −αn2 to find the solution for Tn (t) as Tn (t) = Cn cos(αn ct) + Dn sin(αn ct),
(1.6.18)
24
1 Linear Partial Differential Equations
where Cn and Dn are constants of integration. Combining (1.6.17) with (1.6.18) yields the solution from (1.6.4) as nπx nπct nπct + bn sin sin , (1.6.19) un (x, t) = an cos where an = Cn Bn , bn = Bn Dn are new arbitrary constants and n = 1, 2, 3, . . . . ), are called the These solutions un (x, t), corresponding to eigenvalues αn = ( nπ eigenfunctions. Finally, since the problem is linear, the most general solution is obtained by the principle of superposition in the form ∞ nπct nπct nπx u(x, t) = an cos + bn sin sin , n=1
(1.6.20)
provided the series converges and it is twice continuously differentiable with respect to x and t. The arbitrary coefficients an and bn are determined from the initial conditions (1.6.2ab) which give
nπx an sin , u(x, 0) = f (x) = n=1 ∞ πc nπx . nbn sin ut (x, 0) = g(x) = n=1 ∞
(1.6.21) (1.6.22)
Either by a Fourier series method (see Appendix B) or by direct multiplication of (1.6.21) and (1.6.22) by sin( mπx ) and integrating from 0 to , we can find an and bn as 2 nπx 2 nπx dx, bn = dx, f (x) sin g(x) sin an = 0 nπc 0 (1.6.23ab) in which we have used the result mπx nπx sin (1.6.24) sin dx = δmn , 2 0 where δmn are Kronecker deltas. Thus, (1.6.20) represents the solution where an and bn are given by (1.6.23ab). Hence, the problem is completely solved. We examine the physical significance of the solution (1.6.19) in the context of the free vibration of a string of length . The eigenfunctions nπx nπc , ωn = , (1.6.25) un (x, t) = (an cos ωn t + bn sin ωn t) sin are called the nth normal modes of vibration or the nth harmonic, and ωn represents the discrete spectrum of circular (or radian) frequency or νn = ω2πn = nc 2 , which are called the angular frequencies. The first harmonic (n = 1) is called the fundamental
1.6 The Method of Separation of Variables
25
Fig. 1.4 Several modes of vibration in a string.
harmonic and all other harmonics (n > 1) are called overtones. The frequency of the fundamental mode is given by
πc 1 T∗ , ν1 = . (1.6.26ab) ω1 = 2 ρ Result (1.6.26ab) is considered the fundamental law (or Mersenne law) of a stringed musical instrument. The angular frequency of the fundamental mode of transverse vibration of a string varies as the square root of the tension, inversely as the length, and inversely as the square root of the density. The period of the fundamental mode is 2π = 2 , which is called the fundamental period. Finally, the solution (1.6.20) T1 = ω c 1 describes the motion of a plucked string as a superposition of all normal modes of vibration with frequencies which are all integral multiples (ωn = nω1 or νn = nν1 ) of the fundamental frequency. This is the main reason for the fact that stringed instruments produce more sweet musical sounds (or tones) than drum instruments. In order to describe waves produced in the plucked string with zero initial velocity (ut (x, 0) = 0), we write the solution (1.6.25) in the form nπct nπx un (x, t) = an sin cos , n = 1, 2, 3, . . . . (1.6.27) These solutions are called standing waves with amplitude an sin( nπx ), which vanishes at 2 , , . . . , . n n These are called the nodes of the nth harmonic. The string displays n loops separated by the nodes as shown in Figure 1.4. It follows from elementary trigonometry that (1.6.27) takes the form
nπ nπ 1 un (x, t) = an sin (x − ct) + sin (x + ct) . (1.6.28) 2 x = 0,
26
1 Linear Partial Differential Equations
This shows that a standing wave is expressed as a sum of two progressive waves of equal amplitude traveling in opposite directions. This result is in agreement with the d’Alembert solution. Finally, we can rewrite the solution (1.6.19) of the nth normal modes in the form nπct nπx (1.6.29) cos − εn , un (x, t) = cn sin 1
where cn = (a2n + b2n ) 2 and tan εn = ( abnn ). This solution represents transverse vibrations of the string at any point x and ) and circular frequency ωn = nπc . This at any time t with amplitude cn sin( nπx form of the solution enables us to calculate the kinetic and potential energies of the transverse vibrations. The total kinetic energy (K.E.) is obtained by integrating with respect to x from 0 to , that is,
Kn = K .E . = 0
2 1 ∂un ρ dx, 2 ∂t
(1.6.30)
where ρ is the line density of the string. Similarly, the total potential energy (P.E.) is given by 1 Vn = P .E . = T ∗ 2
0
∂un ∂x
2 dx.
(1.6.31)
Substituting (1.6.29) in (1.6.30) and (1.6.31) gives 2 1 nπc nπct nπx sin2 dx ρ cn sin2 − εn 2 0 1 nπct ρc2 π 2 = (ncn )2 sin2 − εn = ρωn2 c2n sin2 (ωn t − εn ), 4 4
Kn =
(1.6.32)
where ωn = nπc . Similarly, 2 1 ∗ nπcn nπct nπx cos2 cos2 dx T − εn 2 0 nπct 1 π2T ∗ = (ncn )2 cos2 − εn = ρωn2 c2n cos2 (ωn t − εn ). 4 4
Vn =
(1.6.33)
Thus, the total energy of the nth normal modes of vibrations is given by En = Kn + Vn =
1 ρ(ωn cn )2 = const. 4
(1.6.34)
For a given string oscillating in a normal mode, the total energy is proportional to the square of the circular frequency and to the square of the amplitude.
1.6 The Method of Separation of Variables
27
Finally, the total energy of the system is given by E=
∞
En =
n=1
∞ 1 2 2 ω c , ρ 4 n=1 n n
(1.6.35)
which is constant because En = const. Example 1.6.2 (One-Dimensional Diffusion Equation). The temperature distribution u(x, t) in a homogeneous rod of length satisfies the diffusion equation ut = κuxx ,
0 < x < , t > 0,
(1.6.36)
with the boundary and initial conditions t ≥ 0,
(1.6.37ab)
0 ≤ x ≤ ,
(1.6.38)
u(0, t) = 0 = u(, t), u(x, 0) = f (x),
where κ is a diffusivity constant. We assume a separable solution of (1.6.36) in the form u(x, t) = X(x)T (t) = 0.
(1.6.39)
Substituting (1.6.39) in (1.6.36) gives 1 d2 X 1 dT . = 2 X dx κT dt
(1.6.40)
Since the left-hand side depends only on x and the right-hand side is a function of time t only, result (1.6.40) can be true only if both sides are equal to the same constant λ. Thus, we obtain two ordinary differential equations d2 X − λX = 0, dx2
dT − λκT = 0. dt
(1.6.41ab)
For λ ≥ 0, the only solution of the form (1.6.39) consistent with the given boundary conditions is u(x, t) ≡ 0. Hence, for negative λ = −α2 , d2 X + α2 X = 0, dx2
dT + κα2 T = 0, dt
(1.6.42ab)
which admit solutions as X(x) = A cos αx + B sin αx
(1.6.43)
T (t) = C exp −κα2 t ,
(1.6.44)
and
where A, B, and C are constants of integration.
28
1 Linear Partial Differential Equations
The boundary conditions for X(x) are X(0) = 0 = X(),
(1.6.45)
which are used to find A and B in solution (1.6.43). It turns out that A = 0 and B = 0. Hence, sin α = 0,
(1.6.46)
nπ ,
(1.6.47)
which gives the eigenvalues α = αn =
n = 1, 2, 3, . . . .
The value n = 0 is excluded because it leads to a trivial solution. Thus, the eigenfunctions are given by nπx Xn (x) = Bn sin , (1.6.48) where Bn are nonzero constants. With α = αn = nπ , we combine (1.6.44) with (1.6.48) to obtain the solution for un (x, t) as 2 nπ nπx un (x, t) = an exp − , (1.6.49) κt sin where an = Bn Cn is a new constant. Thus, (1.6.47) and (1.6.49) constitute an infinite set of eigenvalues and eigenfunctions. Thus, the most general solution is obtained by the principle of superposition in the form 2 ∞ nπ nπx . (1.6.50) an exp − κt sin u(x, t) = n=1 Now, the initial condition implies that f (x) =
∞ n=1
an sin
nπx ,
which determines an , in view of (1.6.24), as 2 nπx dx. f (x) sin an = 0
(1.6.51)
(1.6.52)
Thus, the final form of the solution is given by 2 ∞ nπx nπ 2 nπx f x sin κt sin u(x, t) = dx exp − . 0 n=1 (1.6.53)
1.6 The Method of Separation of Variables
29
It follows from the series solution (1.6.53) that the series satisfies the given boundary and initial conditions. It also satisfies equation (1.6.36) because the series is convergent for all x(0 ≤ x ≤ ) and t ≥ 0 and can be differentiated term by term. Physically, the temperature distribution decays exponentially with time t. This shows a striking contrast to the wave equation, whose solution oscillates in time t. The time 2 ) which is directly proportional to 2 scale of decay for the nth mode is Td ∼ κ1 ( nπ and inversely proportional to the thermal diffusivity. The method of separation of variables is applicable to the wave equation and the diffusion equation, and also to problems involving Laplace’s equation and other equations in two or three dimensions with a wide variety of initial and boundary conditions. We consider the following examples. Example 1.6.3 (Two-Dimensional Diffusion Equation). We consider ut = κ(uxx + uyy ), 0 < x < a, 0 < y < b, t > 0, u(x, y, t) = f (x, y) at t = 0, u(x, y, t) = 0
on ∂D,
(1.6.54) (1.6.55) (1.6.56)
where ∂D is the boundary of the rectangle defined by 0 ≤ x ≤ a, 0 ≤ y ≤ b. The method here is precisely the same as in the previous examples except that we seek a solution of (1.6.54) in the form u(x, y, z) = S(x, y)T (t) = 0,
(1.6.57)
so that S and T satisfy the equations ∂ 2S ∂2S + + λS = 0, 2 ∂x ∂y 2 ∂T + κλT = 0. ∂t
(1.6.58) (1.6.59)
For λ ≤ 0, the separable solution (1.6.57) with the given boundary data leads only to a trivial solution u(x, y, t) ≡ 0. Hence, for positive λ, we solve (1.6.58), (1.6.59) subject to the given boundary and initial conditions. Equation (1.6.58) is an elliptic equation, and here we seek a solution S(x, y) which satisfies the boundary conditions S(0, y) = 0 = S(a, y) for 0 ≤ y ≤ b, S(x, 0) = 0 = S(x, b) for 0 ≤ x ≤ a.
(1.6.60) (1.6.61)
We also seek a separable solution of (1.6.58) in the form S(x, y) = X(x)Y (y) = 0 and find that X(x) and Y (y) satisfy the equation Y X + + λ = 0, X Y
(1.6.62)
30
that is,
1 Linear Partial Differential Equations
Y X = −μ = − +λ . X Y
(1.6.63)
Or X + μX = 0,
(1.6.64a)
Y + (λ − μ)Y = 0.
(1.6.64b)
These equations have to be solved with the boundary conditions X(0) = 0 = X(a),
(1.6.65a)
Y (0) = 0 = Y (b).
(1.6.65b)
The eigenvalue problem (1.6.64a), (1.6.64b) with (1.6.65a), (1.6.65b) gives the eigenvalues 2 mπ μm = , (1.6.66) a and the corresponding eigenfunctions
mπx Xm (x) = Am sin , a
(1.6.67)
when m = 1, 2, 3, . . . . Thus, equation (1.6.64b) becomes Y + (λ − μm )Y = 0,
(1.6.68)
which has to be solved with (1.6.65b). This is another eigenvalue problem similar to that already considered and leads to the eigenvalues 2 nπ (1.6.69) λn − μ m = b and the corresponding eigenfunctions
nπy Yn (y) = Bn sin , b
(1.6.70)
where n = 1, 2, 3, . . . . In other words, the solution of equation (1.6.58) becomes mπx nπy Smn (x, y) = Xm (x)Yn (y) = Amn sin sin , (1.6.71) a b where Amn = Am Bn are constants together with the eigenvalues 2 2 nπ m n2 π2 , = + λmn = μm + b a2 b2 where m = 1, 2, 3, . . . and n = 1, 2, 3, . . . .
(1.6.72)
1.6 The Method of Separation of Variables
31
With λmn as eigenvalues, we solve (1.6.59) to obtain (1.6.73)
Tmn (t) = Bmn exp(−λmn κt), where Bmn are integrating constants. Finally, the solution (1.6.57) can be expressed as a double series
∞ ∞
mπx amn sin u(x, y, t) = a m=1 n=1
nπy sin exp(−λmn κt), b
(1.6.74)
where amn are constants to be determined from the initial condition so that ∞ ∞ mπx nπy amn sin f (x, y) = u(x, y, 0) = sin . (1.6.75) a b m=1 n=1 To find constants amn , we multiply (1.6.75) by sin( rπx a ) and integrate the result with respect to x from 0 to a with fixed y, so that a ∞ nπy mπx a = dx. amn sin f (x, y) sin 2 n=1 b a 0 The right-hand side is a function of y and is set equal to g(y), so that a mπx f (x, y) sin dx = g(y). a 0
(1.6.76)
(1.6.77)
Then, the coefficients amn (m fixed) in (1.6.76) are found by multiplying it by sin( nπy b ) and integrating with respect to y from 0 to b, so that
b nπy ab amn = dy, g(y) sin 4 b 0
(1.6.78)
whence amn =
4 ab
a
b
f (x, y) sin 0
0
mπx a
sin
nπy b
dx dy.
(1.6.79)
Thus, the solution of the problem is given by (1.6.74) where amn is represented by (1.6.79). The method of construction of the solution shows that the initial and boundary conditions are satisfied by the solution. Moreover, the uniform convergence of the double series justifies differentiation of the series, and this, in turn, permits us to verify the solution by direct substitution in the original diffusion equation (1.6.54). Example 1.6.4 (Dirichlet’s Problem for a Circle). We consider the Laplace equation in cylindrical polar coordinates (r, θ, z) as 1 1 urr + ur + 2 uθθ = 0, r r
0 < r < a, 0 ≤ θ < 2π,
(1.6.80)
32
1 Linear Partial Differential Equations
with the boundary condition u(a, θ) = f (θ)
for all θ.
(1.6.81)
According to the method of separation of variables, we seek a solution in the form u(r, θ) = R(r)Θ(θ) = 0.
(1.6.82)
Substituting this solution in equation (1.6.80) gives r2
R Θ R +r =− = λ. R R Θ
Hence, r 2 R + rR − λR = 0 and
Θ + λΘ = 0.
(1.6.83a) (1.6.83b)
For Θ(θ), we naturally require periodic boundary conditions Θ(θ + 2π) = Θ(θ)
for − ∞ < z < ∞.
(1.6.84)
Due to the periodicity condition, for λ < 0, the solution (1.6.82) leads to a trivial solution. So, there are two cases: (i) λ = 0 and (ii) λ > 0. For case (i), we have the solution u(r, θ) = (A + B log r)(Cθ + D).
(1.6.85)
Since log r is singular at r = 0, hence, B = 0. For u to be periodic with period 2π, C = 0. Hence, the solution u must be constant for λ = 0. For λ > 0, the solution of equation (1.6.83b) is √ √ (1.6.86) Θ(θ) = A cos λθ + B sin λθ. √ Since Θ(θ) is periodic with period 2π, λ must be an integer n so that λ = n2 , n = 1, 2, 3, . . . . Thus, solution (1.6.86) becomes Θ(θ) = A cos nθ + B sin nθ.
(1.6.87)
The equation (1.6.83a) is the Euler equation with λ = n2 . It gives solutions of the form R(r) = rα = 0 so that (1.6.83a) gives α(α − 1) + α − n2 rα = 0, whence α = ±n. Thus, the solution for R(r) is given by R(r) = Crn + Dr−n .
(1.6.88)
Since R(r) → ∞ as r → 0 because of the term r−n , we get D = 0. Thus, the solution (1.6.82) reduces to
1.6 The Method of Separation of Variables
u(r, θ) = Crn (A cos nθ + B sin nθ).
33
(1.6.89)
By the superposition principle, the solution of the Laplace equation within a circular region including the origin r = 0 is u(r, θ) =
∞ 1 a0 + rn (an cos nθ + bn sin nθ), 2 n=1
(1.6.90)
where a0 , an , and bn are constants to be determined from the boundary conditions, and the first term 12 a0 represents the constant solution for λ = 0 (n = 0). Finally, using the boundary condition (1.6.81), we derive f (θ) = u(a, θ) =
∞ 1 a0 + an (an cos nθ + bn sin nθ). 2 n=1
(1.6.91)
This is exactly the Fourier series representation for f (θ), and hence, the coefficients are given by 1 an = πan bn =
1 πan
2π
f (φ) cos nφ dφ,
n = 0, 1, 2, 3, . . . ,
f (φ) sin nφ dφ,
n = 1, 2, 3, . . . .
0 2π 0
Substituting the values for an and bn into (1.6.91) yields the solution 1 u(r, θ) = 2π
2π
f (φ) dφ 0
∞ n 2π 1 r (cos nθ cos nφ + sin nθ sin nφ)f (φ) dφ π n=1 a 0 2π ∞ n r 1 f (φ) 1 + 2 cos n(θ − φ) dφ, (1.6.92) = 2π 0 a n=1
+
where the term inside the set of braces in the above integral can be summed by writing it as a geometric series, that is, 1+
∞ n r n=1
a
exp in(θ − φ) +
∞ n r n=1
a
exp −in(θ − φ)
r exp{i(θ − φ)} r exp{−i(θ − φ)} =1+ + a − r exp{i(θ − φ)} a − r exp{−i(θ − φ)} (a2 − r2 ) . = 2 a − 2ar cos(θ − φ) + r2 Thus, the final form of the solution is
34
1 Linear Partial Differential Equations
u(r, θ) =
1 2π
2π
a2
0
(a2 − r2 )f (φ) dφ . − 2ar cos(θ − φ) + r2
(1.6.93)
This formula is known as Poisson’s integral formula representing the solution of the Laplace equation within the circle of radius a in terms of values prescribed on the circle. It has several important consequences. First, we set r = 0 and θ = 0 in (1.6.93) to obtain 2π 1 u(0, 0) = f (φ) dφ. (1.6.94) 2π 0 This states that the value of u at the center is equal to the mean value of u on the boundary of the circle. This is called the mean value property. We rewrite (1.6.93) in the form 2π P (r, θ − φ)f (φ) dφ, (1.6.95) u(r, θ) = 0
where P (r, θ − φ) is called the Poisson kernel given by P (r, θ − φ) =
1 (a2 − r 2 ) , · 2 2π a − 2ar cos(θ − φ) + r 2
(1.6.96)
which is zero for r = a but θ = φ. Further, 2π f (θ) = lim u(r, θ) = lim P (r, θ − φ) f (φ) dφ, r→a−
0
r→a−
which implies that lim P (r, θ − φ) = δ(θ − φ),
r→a−
(1.6.97)
where δ(x) is the Dirac delta function.
1.7 Fourier Transforms and Initial Boundary-Value Problems The Fourier transform of u(x, t) with respect to x ∈ R is denoted by F{u(x, t)} = U (k, t) and is defined by the integral ∞ 1 F u(x, t) = U (k, t) = √ e−ikx u(x, t) dx, (1.7.1) 2π −∞ where k is real and is called the transform variable. The inverse Fourier transform, denoted by F −1 {U (k, t)} = u(x, t), is defined by ∞ 1 −1 U (k, t) = u(x, t) = √ eikx U (k, t) dk. (1.7.2) F 2π −∞ The basic properties of the Fourier transforms including the convolution are discussed in Appendix B.
1.7 Fourier Transforms and Initial Boundary-Value Problems
35
Example 1.7.1. 2 k 1 2 , a > 0, = √ exp − (a) F exp −ax 4a 2a 2 a (b) F exp −a|x| = · , a > 0. π (k2 + a2 )
(1.7.3) (1.7.4)
If u(x, t) → 0, as |x| → ∞, then ∞ ∂u ∂u 1 e−ikx F =√ dx, ∂x ∂x 2π −∞ which, by integrating by parts, is ∞ ik 1 −ikx u(x, t) −∞ + √ = √ e 2π 2π = ikF u(x, t) = ikU (k, t).
∞
e−ikx u(x, t) dx
−∞
(1.7.5) k
Similarly, if u(x, t) is continuously n times differentiable and ∂∂xuk → 0, as |x| → ∞ for k = 1, 2, 3, . . . , (n − 1), then n ∂ u = (ik)n F u(x, t) = (ik)n U (k, t). (1.7.6) F n ∂x It also follows from the definition (1.7.1) that 2 dU ∂ u ∂u d2 U = , F ,..., F = ∂t dt ∂t2 dt2
∂nu F ∂tn
=
dn U . dtn
(1.7.7)
The definition of the Fourier transform (1.7.1) shows that a sufficient condition for u(x, t) to have a Fourier transform is that u(x, t) is absolutely integrable on −∞ < x < ∞. This existence condition is too strong for many practical applications. Many simple functions, such as a constant function, sin ωx, and xn H(x), do not have Fourier transforms, even though they occur frequently in applications. The above definition of the Fourier transform has been extended for a more general class of functions to include the above and other functions. We simply state the fact that there is a sense, useful in practical applications, in which the above stated functions and many others do have Fourier transforms. The following are examples of such functions and their Fourier transforms: 2 sin ak , (1.7.8) F H a − |x| = π k where H(x) is the Heaviside unit step function, 1 F δ(x − a) = √ exp(−iak), 2π
(1.7.9)
36
1 Linear Partial Differential Equations
where δ(x − a) is the Dirac delta function, and
π 1 + δ(k) exp(−iak). F H(x − a) = 2 iπk
(1.7.10)
Theorem 1.7.1 (Convolution Theorem). If F{f (x)} = F (k) and F{g(x)} = G(k), then F f (x) ∗ g(x) = F (k)G(k).
(1.7.11)
F −1 F (k)G(k) = f (x) ∗ g(x),
(1.7.12)
Or equivalently,
where f (x) ∗ g(x) is called the convolution of two integrable functions f (x) and g(x) and is defined by 1 f (x) ∗ g(x) = √ 2π
∞ −∞
f (x − ξ)g(ξ) dξ.
(1.7.13)
Hence, the result (1.7.12) can also be written as
∞
−∞
f (x − ξ)g(ξ) dξ =
∞
eikx F (k)G(k) dk.
(1.7.14)
−∞
It can easily be verified that the convolution has the following simple properties (see Appendix B): f (x) ∗ g(x) = g(x) ∗ f (x) (commutative), √ √ f (x) ∗ 2πδ(x) = f (x) = 2πδ(x) ∗ f (x) (identity).
(1.7.15) (1.7.16)
The Fourier transforms are very useful in solving a wide variety of initial boundaryvalue problems governed by linear partial differential equations. The following examples of applications illustrate the method of Fourier transforms. Example 1.7.2 (The Cauchy Problem for the Linear Wave Equation). d’Alembert solution of the initial-value problem for the wave equation utt = c2 uxx ,
−∞ < x < ∞, t > 0,
Obtain the
(1.7.17)
with the initial data u(x, 0) = f (x),
ut (x, 0) = g(x),
−∞ < x < ∞.
(1.7.18ab)
Application of the Fourier transform F{u(x, t)} = U (k, t) to this system gives
1.7 Fourier Transforms and Initial Boundary-Value Problems
37
d2 U + c2 k 2 U = 0, dt2 dU U (k, 0) = F (k), = G(k). dt t=0 The solution of the transformed system is U (k, t) = Aeickt + Be−ickt , where A and B are constants of integration to be determined from the transformed 1 data A + B = F (k) and A − B = ikc G(k). Solving for A and B, we obtain the solution U (k, t) =
G(k) ickt 1 F (k) eickt + e−ickt + e − e−ickt . 2 2ick
Thus, the inverse Fourier transform of (1.7.19) yields the solution
∞ 1 1 √ u(x, t) = F (k) eik(x+ct) + eik(x−ct) dk 2 2π −∞
∞ 1 1 G(k) ik(x+ct) √ e + − eik(x−ct) dk . 2c 2π −∞ ik
(1.7.19)
(1.7.20)
We use the following results: ∞ 1 √ f (x) = F F (k) = eikx F (k) dk, 2π −∞ ∞ 1 −1 √ G(k) = g(x) = F eikx G(k) dk, 2π −∞ −1
to obtain the solution in the form 1 u(x, t) = f (x − ct) + f (x + ct) + 2 1 = f (x − ct) + f (x + ct) + 2 1 = f (x − ct) + f (x + ct) + 2
∞ x+ct 1 1 √ G(k) dk eikξ dξ 2c 2π −∞ x−ct
x+ct ∞ 1 1 ikξ dξ √ e G(k) dk 2c x−ct 2π −∞ x+ct 1 g(ξ) dξ. (1.7.21) 2c x−ct
This is the well-known d’Alembert solution of the wave equation. In particular, if f (x) = exp(−x2 ) and g(x) = 0, then the d’Alembert solution (1.7.21) reduces to u(x, t) =
1 exp −(x − ct)2 + exp −(x + ct)2 . 2
(1.7.22)
This shows that the initial wave profile breaks up into two identical traveling waves of half the amplitude moving in opposite directions with speed c.
38
1 Linear Partial Differential Equations
On the other hand, if f (x) = 0 and g(x) = δ(x), the d’Alembert solution (1.7.21) becomes x+ct 1 1 x+ct δ(ξ) dξ = H (ξ) dξ u(x, t) = 2c x−ct 2c x−ct 1 H(x + ct) − H(x − ct) = 2c 1 1 if |x| < ct, = (1.7.23) 2c 0 if |x| > ct > 0 1 = H c 2 t2 − x 2 . (1.7.24) 2c Example 1.7.3 (The Linearized Korteweg–de Vries Equation). The linearized Korteweg–de Vries (KdV) equation for the free surface elevation η(x, t) in inviscid water of constant depth h is ηt + cηx +
ch2 ηxxx = 0, 6
−∞ < x < ∞, t > 0,
(1.7.25)
√ where c = gh is the shallow water speed. We solve equation (1.7.25) with the initial condition η(x, 0) = f (x),
−∞ < x < ∞.
(1.7.26)
Application of the Fourier transform F{η(x, t)} = E(k, t) to the KdV system gives the solution for E(k, t) in the form 2 2
k h E(k, t) = F (k) exp ikct −1 . 6 The inverse transform gives 2
∞ 1 cth η(x, t) = √ k 2 dk. F (k) exp ik (x − ct) + 6 2π −∞ In particular, if f (x) = δ(x), then (1.7.27) reduces to the Airy integral 2
1 ∞ cth η(x, t) = k 3 dk, cos k(x − ct) + π 0 6 which, in terms of the Airy function, 2 − 13 2 − 13
cth cth = Ai (x − ct) , 2 2 where the Airy function Ai (az) is defined by ∞ k3 1 cos kz + 3 dk. Ai (az) = πa 0 3a
(1.7.27)
(1.7.28)
(1.7.29)
(1.7.30)
1.7 Fourier Transforms and Initial Boundary-Value Problems
39
Example 1.7.4 (Dirichlet’s Problem in the Half Plane). We consider the solution of the Laplace equation in the half plane −∞ < x < ∞, y ≥ 0,
uxx + uyy = 0,
(1.7.31)
with the boundary conditions u(x, 0) = f (x), −∞ < x < ∞, u(x, y) → 0 as |x| → ∞, y → ∞.
(1.7.32) (1.7.33)
We apply the Fourier transform with respect to x to the system (1.7.31)–(1.7.33) to obtain d2 U − k 2 U = 0, dy 2 and U (k, y) → 0
U (k, 0) = F (k),
(1.7.34) as y → ∞.
Thus, the solution of this transformed system is U (k, y) = F (k) exp −|k|y .
(1.7.35ab)
(1.7.36)
Application of the Convolution Theorem 1.7.1 gives the solution ∞ 1 f (ξ)g(x − ξ) dξ, u(x, y) = √ 2π −∞
(1.7.37)
where g(x) = F
−1
e
−|k|y
=
y 2 . 2 π (x + y 2 )
Consequently, the solution (1.7.37) becomes f (ξ) dξ y ∞ u(x, y) = , π −∞ (x − ξ)2 + y 2
y > 0.
(1.7.38)
(1.7.39)
This is the well-known Poisson integral formula in the half plane. It is noted that
∞ y 1 dξ · lim u(x, y) = f (ξ) lim y→0+ y→0+ π (x − ξ)2 + y 2 −∞ ∞ = f (ξ)δ(x − ξ) dξ, (1.7.40) −∞
where Cauchy’s definition of the delta function is used, that is, δ(x − ξ) = lim+ y→0
1 y · . π (x − ξ)2 + y 2
(1.7.41)
40
1 Linear Partial Differential Equations
Fig. 1.5 Isothermal curves representing a family of circular arcs.
This may be recognized as a solution of the Laplace equation for a dipole source at (x, y) = (ξ, 0). In particular, when (1.7.42) f (x) = T0 H a − |x| , the solution (1.7.39) reduces to
yT0 a dξ π −a (x − ξ)2 + y 2
x+a x−a T0 tan−1 − tan−1 = π y y 2ay T0 . tan−1 2 = π x + y 2 − a2
u(x, y) =
(1.7.43)
The curves in the upper half plane, for which the steady-state temperature is constant, are known as isothermal curves. In this case, these curves represent a family of circular arcs x2 + y 2 − αy = a2
(1.7.44)
with centers on the y-axis and fixed end points on the x-axis at x = ±a, as shown in Figure 1.5. Another special case deals with f (x) = δ(x). This solution for this case follows from (1.7.39) and is given by y δ(ξ) dξ 1 y ∞ . = u(x, t) = π −∞ (x − ξ)2 + y 2 π (x2 + y 2 )
(1.7.45)
(1.7.46)
1.7 Fourier Transforms and Initial Boundary-Value Problems
41
Further, we can readily reduce the solution of the Neumann problem in the half plane from the solution of the Dirichlet problem. Example 1.7.5 (Neumann’s Problem in the Half Plane). Find a solution of the Laplace equation uxx + uyy = 0,
−∞ < x < ∞, y > 0,
(1.7.47)
with the Neumann boundary condition uy (x, 0) = f (x),
−∞ < x < ∞.
(1.7.48)
Condition (1.7.48) specifies the normal derivative on the boundary, and, physically, it describes the fluid flow or heat flux at the boundary. We define a new function v(x, y) = uy (x, y) so that y v(x, η) dη, (1.7.49) u(x, y) = where an arbitrary constant can be added to the right-hand side. Clearly, the function v satisfies the Laplace equation ∂2v ∂ 2 uy ∂ 2 uy ∂ ∂ 2v (uxx + uyy ) = 0, + = + = ∂x2 ∂y 2 ∂x2 ∂y 2 ∂y with the boundary condition v(x, 0) = uy (x, 0) = f (x)
for − ∞ < x < ∞.
Thus, v(x, y) satisfies the Laplace equation with the Dirichlet condition on the boundary. Obviously, the solution is given by (1.7.39), that is, y ∞ f (ξ) dξ . (1.7.50) v(x, y) = π −∞ (x − ξ)2 + y 2 Then, the solution u(x, y) can be obtained from (1.7.49) in the form ∞ y 1 y f (ξ) dξ v(x, η) dη = η dη u(x, y) = 2 2 π −∞ (x − ξ) + η ∞ y 1 η dη = f (ξ) dξ , y>0 π −∞ (x − ξ)2 + η 2 ∞ 1 = f (ξ) log (x − ξ)2 + y 2 dξ, 2π −∞
(1.7.51)
where an arbitrary constant can be added to this solution. In other words, the solution of any Neumann problem is uniquely determined up to an arbitrary constant.
42
1 Linear Partial Differential Equations
Example 1.7.6 (The Cauchy Problem for the Diffusion Equation). We consider the initial-value problem for a one-dimensional diffusion equation with no sources or sinks: −∞ < x < ∞, t > 0,
ut = κuxx ,
(1.7.52)
where κ is a diffusivity constant with the initial condition u(x, 0) = f (x),
−∞ < x < ∞.
(1.7.53)
We solve this problem using the Fourier transform in the space variable x defined by (1.7.1). Application of this transform to (1.7.52), (1.7.53) gives Ut = −κk 2 U, U (k, 0) = F (k).
(1.7.54)
t > 0,
(1.7.55)
The solution of the transformed system is U (k, t) = F (k) exp −κk2 t .
(1.7.56)
The inverse Fourier transform gives the solution ∞ 1 F (k) exp ikx − κk2 t dk, u(x, t) = √ 2π −∞ which is, by the Convolution Theorem 1.7.1, ∞ 1 u(x, t) = √ f (ξ)g(x − ξ) dξ, 2π −∞
(1.7.57)
where 2 x2 1 , g(x) = F −1 e−κk t = √ exp − 4κt 2κt
by (1.7.3).
Thus, solution (1.7.57) becomes u(x, t) = √
1 4πκt
(x − ξ)2 dξ. f (ξ) exp − 4κt −∞ ∞
(1.7.58)
The integrand involved in the solution consists of the initial value f (x) and the Green’s function (or the fundamental solution) G(x − ξ, t) of the diffusion equation for the infinite interval:
(x − ξ)2 1 . (1.7.59) exp − G(x − ξ, t) = √ 4κt 4πκt So, in terms of G(x − ξ, t), solution (1.7.58) can be written as
1.7 Fourier Transforms and Initial Boundary-Value Problems
∞
u(x, t) = −∞
f (ξ)G(x − ξ, t) dξ,
43
(1.7.60)
so that, in the limit as t → 0+, this formally becomes ∞ f (ξ) lim G(x − ξ, t) dξ. u(x, 0) = f (x) = t→0+
−∞
The limit of G(x − ξ, t), as t → 0+, represents the Dirac delta function
1 (x − ξ)2 δ(x − ξ) = lim √ . exp − t→0+ 4κt 4πκt
(1.7.61)
It is important to point out that the integrand in (1.7.60) consists of the initial temperature distribution f (x) and the Green’s function G(x − ξ, t) which represents the temperature response along the rod at time t due to an initial unit impulse of heat at x = ξ. The physical meaning of the solution (1.7.60) is that the initial temperature distribution f (x) is decomposed into a spectrum of impulses of magnitude f (ξ) at each point x = ξ to form the resulting temperature f (ξ)G(x − ξ, t). According to the linear superposition principle (1.3.6), the resulting temperature is integrated to find solution (1.7.60). We make the change of variable ξ−x √ = ζ, 2 κt
dξ dζ = √ 2 κt
to express solution (1.7.58) in the form ∞ √ 1 u(x, t) = √ f x + 2 κtζ exp −ζ 2 dζ. π −∞
(1.7.62)
The integral solution (1.7.62) or (1.7.58) is called the Poisson integral representation of the temperature distribution. This integral is convergent for all time t > 0, and the integrals obtained from (1.7.62) by differentiation under the integral sign with respect to x and t are uniformly convergent in the neighborhood of the point (x, t). Hence, the solution u(x, t) and its derivatives of all orders exist for t > 0. Finally, we consider two special cases: (a) f (x) = δ(x) and (b) f (x) = T0 H(x), where T0 is a constant. For case (a), the solution (1.7.58) reduces to
∞ 1 (x − ξ)2 u(x, t) = √ dξ δ(ξ) exp − 4κt 4πκt −∞ 1 x2 . = √ exp − 4κt 4πκt
(1.7.63)
This is usually called Green’s function or the fundamental solution of the diffusion equation and is shown in Figure 1.6.
44
1 Linear Partial Differential Equations
Fig. 1.6 √ The temperature distribution u(x, t) due to a point source for different values of τ = 2 κt.
At any time t, the √ solution u(x, t) is Gaussian. The peak height√of u(x, t) deκt, whereas the width of the solution (x ∼ κt) increases creases inversely with √ with κt. In fact, the initially sharply peaked profile is gradually smoothed out as t → ∞ under the action of diffusion. These are remarkable features for diffusion phenomena. For case (b), the initial data is discontinuous. In this case, the solution is
∞ (x − ξ)2 T0 dξ. (1.7.64) exp − u(x, t) = √ 4κt 2 πκt 0 √ , we can express solution (1.7.64) in the Introducing the change of variable η = 2ξ−x κt form ∞ x T0 T0 −η 2 erfc − √ e dη = u(x, t) = √ 2 π −x 2 κt √ 2 κt
T0 x = 1 + erf √ . (1.7.65) 2 2 κt
This shows that, at t = 0, the solution coincides with the initial data u(x, 0) = T0 . The graph of T10 u(x, t) against x is shown in Figure 1.7. As t increases, the discontinuity √ is gradually smoothed out, whereas the width of the transition zone increases as κt. Example 1.7.7 (The Schrödinger Equation in Quantum Mechanics). The time-dependent Schrödinger equation for a particle of mass m is
2 2 (1.7.66) ∇ ψ = Hψ, iψt = V (x) − 2m where h(= 2π) is the Planck constant, ψ(x, t) is the wave function, V (x) is the ∂2 ∂2 ∂2 potential, ∇2 = ∂x 2 + ∂y 2 + ∂z 2 is the three-dimensional Laplacian, and H is called the Hamiltonian.
1.7 Fourier Transforms and Initial Boundary-Value Problems
45
Fig. 1.7 √ The temperature distribution due to discontinuous initial data for different values of τ = 2 κt = 0, 1, 2, 5.
If V (x) = const. = V , we seek a plane wave solution of the form ψ(x, t) = A exp i(κ · x − ωt) ,
(1.7.67)
where A is a constant amplitude, κ = (k, l, m) is the wavenumber vector, and ω is the frequency. Substituting this solution into (1.7.66), we conclude that this solution is possible provided the following relation is satisfied: i(−iω) = V −
2 (iκ)2 , 2m
κ2 = k2 + l2 + m2 ,
or ω = V +
2 κ2 . 2m
(1.7.68)
This is called the dispersion relation for the de Broglie wave and shows that the 2 sum of the potential energy V and the kinetic energy (k) 2m equals the total energy ω. Further, the kinetic energy K .E . =
p2 1 (κ)2 = , 2m 2m
(1.7.69)
where p = κ is the momentum of the particle. In the one-dimensional case, the group velocity is Cg =
κ p mv ∂ω = = = = v. ∂k m m m
(1.7.70)
This shows that the group velocity is equal to the classical particle velocity v.
46
1 Linear Partial Differential Equations
We now use the Fourier transform method to solve the one-dimensional Schrödinger equation for a free particle (V ≡ 0), that is, 2 ψxx , −∞ < x < ∞, t > 0, 2m ψ(x, 0) = ψ0 (x), −∞ < x < ∞,
(1.7.72)
ψ(x, t) → 0
(1.7.73)
iψt = −
as |x| → ∞, t > 0.
(1.7.71)
Application of the Fourier transform to (1.7.71)–(1.7.73) gives ik 2 Ψ, 2m Ψ (k, 0) = Ψ0 (k). Ψt = −
(1.7.74) (1.7.75)
The solution of this transformed system is Ψ (k, t) = Ψ0 (k) exp −iαk2 t ,
α=
. 2m
(1.7.76)
This solution is similar to (1.7.56) with κ = iα so that the inverse Fourier transform gives the formal solution for the wave function ψ(x, t) similar to (1.7.58) in the form
∞ 1 (x − ξ)2 dξ (1.7.77) ψ(x, t) = √ ψ0 (ξ) exp − 4iαt 4πiαt −∞
(1 − i) ∞ i(x − ξ)2 = √ ψ0 (ξ, 0) exp dξ. (1.7.78) 4αt 8παt −∞ This is the integral solution of the one-dimensional Schrödinger equation.
1.8 Multiple Fourier Transforms and Partial Differential Equations Definition 1.8.1. Under the assumptions on f (x) similar to those made for the one-dimensional case, the multiple Fourier transform of f (x), where x = (x1 , x2 , . . . , xn ) is the n-dimensional vector, is defined by ∞ ∞ 1 κ · x) f (x) dx, κ) = · · · exp −i(κ (1.8.1) F f (x) = F (κ n (2π) 2 −∞ −∞ where κ = (k1 , k2 , . . . , kn ) is the n-dimensional transform vector and κ · x = (k1 x1 + k2 x2 + · · · + kn xn ). The inverse Fourier transform is similarly defined by ∞ ∞ 1 −1 κ · x) F (κ κ) dκ κ. κ) = f (x) = F ··· exp i(κ F (κ n 2 (2π) −∞ −∞
(1.8.2)
1.8 Multiple Fourier Transforms and Partial Differential Equations
In particular, the double Fourier transform is defined by ∞ ∞ 1 κ · r) f (x, y) dx dy, exp −i(κ F f (x, y) = F (k, ) = 2π −∞ −∞ where r = (x, y) and κ = (k, ). The inverse double Fourier transform is given by ∞ ∞ 1 κ · r) F (k, ) dk d. exp i(κ F F (k, ) = f (x, y) = 2π −∞ −∞
47
(1.8.3)
(1.8.4)
Similarly, the three-dimensional Fourier transform and its inverse are defined by the integrals F f (x, y, z) = F (k, , m) ∞ ∞ ∞ 1 κ · r) f (x, y, z) dx dy dz, = exp −i(κ 3 (2π) 2 −∞ −∞ −∞ (1.8.5) F −1 F (k, , m) = f (x, y, z) ∞ ∞ ∞ 1 κ · r) F (k, , m) dk d dm. = exp i(κ 3 (2π) 2 −∞ −∞ −∞ (1.8.6) The operational properties of these multiple Fourier transforms are similar to those of the one-dimensional case. In particular, results (1.7.5) and (1.7.6) relating the Fourier transforms of derivatives to the Fourier transforms of given functions are also valid for the higher dimensional case. In higher dimensions, they are applied to the transforms of partial derivatives of f (x) under the assumptions that f (x1 , x2 , . . . , xn ) and its partial derivatives vanish at infinity. We illustrate the multiple Fourier transform method by the following examples of applications. Example 1.8.1 (The Dirichlet Problem for the Three-Dimensional Laplace Equation in the Half-Space). The boundary-value problem for u(x, y, z) satisfies the following equation and boundary conditions: ∇2 u ≡ uxx + uyy + uzz = 0, −∞ < x, y < ∞, z > 0, u(x, y, 0) = f (x, y), −∞ < x, y < ∞, u(x, y, z) → 0, as r = x2 + y 2 + z 2 → ∞.
(1.8.7) (1.8.8) (1.8.9)
We apply the double Fourier transform defined by (1.8.3) to the system (1.8.7)– (1.8.9) which reduces to d2 U − κ2 U = 0 for z > 0, dz 2 U (k, , 0) = F (k, ). Thus, the solution of this transformed problem is
48
1 Linear Partial Differential Equations
U (k, , z) = F (k, ) exp −|κ|z = F (k, )G(k, ), where κ = (k, ) and G(k, ) = exp(−|κ|z), so that z g(x, y) = F −1 exp −|κ|z = 3 . 2 (x + y 2 + z 2 ) 2
(1.8.10)
(1.8.11)
Applying the Fourier Convolution Theorem to (1.8.10) gives the formal solution ∞ ∞ 1 f (ξ, η)g(x − ξ, y − η, z) dξ dη u(x, y, z) = 2π −∞ −∞ ∞ ∞ z f (ξ, η) dξ dη = . (1.8.12) 2π −∞ −∞ [(x − ξ)2 + (y − η)2 + z 2 ] 32 Example 1.8.2 (The Two-Dimensional Diffusion Equation). We solve the two-dimensional diffusion equation ut = K∇2 u,
−∞ < x, y < ∞, t > 0,
(1.8.13)
with the initial conditions u(x, y, 0) = f (x, y), −∞ < x, y < ∞, u(x, y, 0) → 0, as r = x2 + y 2 → ∞,
(1.8.14) (1.8.15)
where K is the diffusivity constant. The double Fourier transform of u(x, y, t), defined by (1.8.3), is used to reduce the system (1.8.13)–(1.8.15) into the form dU = −κ2 KU, dt U (k, , 0) = F (k, ).
t > 0,
The solution of this system is
U (k, , t) = F (k, ) exp −tKκ2 = F (k, )G(k, ),
where
(1.8.16)
G(k, ) = exp −Kκ2 t ,
so that g(x, y) = F −1 exp −Kκ2 t =
2 1 x + y2 . exp − 2Kt 4Kt
(1.8.17)
Finally, applying the Convolution Theorem to (1.8.16) gives the formal solution
∞ ∞ (x − ξ)2 + (y − η)2 1 f (ξ, η) exp − dξ dη, u(x, y, t) = 4πKt −∞ −∞ 4Kt (1.8.18) or equivalently,
1.8 Multiple Fourier Transforms and Partial Differential Equations
u(x, y, t) =
1 4πKt
∞
−∞
|r − r |2 f r exp − dr , 4Kt −∞
49
∞
(1.8.19)
where r = (ξ, η). √ We make the change of variable (r − r ) = 4KtR to reduce (1.8.19) into the form ∞ ∞ √ 1 (1.8.20) u(x, y, t) = √ f r + 4KtR exp −R2 dR. π 4Kt −∞ −∞ Similarly, the formal solution of the initial-value problem for the three-dimensional diffusion equation ut = K(uxx + uyy + uzz ), −∞ < x, y, z < ∞, t > 0, u(x, y, z, 0) = f (x, y, z), −∞ < x, y, z < ∞,
(1.8.21) (1.8.22)
is given by u(x, y, z, t) =
1 3
(4πKt) 2
r2 dξ dη dζ, f (ξ, η, ζ) exp − 4Kt −∞ ∞
(1.8.23)
where r 2 = (x − ξ)2 + (y − η)2 + (z − ζ)2 . Or equivalently, u(x, y, z, t) =
|r − r |2 dξ dη dζ, f r exp − 4Kt −∞
1 3
(4πKt) 2
∞
(1.8.24)
where r = (x, y, z) and r = (ξ, η, ζ). √ Making the change of variable r − r = 4tKR, solution (1.8.24) reduces to the form ∞ √ 1 u(x, y, z, t) = 3 f r + 4KtR exp −R2 dR. (1.8.25) 2 π 4Kt −∞ This is known as the Fourier integral solution. Example 1.8.3 (The Cauchy Problem for the Two-Dimensional Wave Equation). The initial-value problem for the wave equation in two dimensions is governed by utt = c2 (uxx + uyy ),
−∞ < x, y < ∞, t > 0,
(1.8.26)
with the initial data u(x, y, 0) = 0,
ut (x, y, 0) = f (x, y),
−∞ < x, y < ∞,
(1.8.27ab)
where c is a constant. We assume that u(x, y, t) and its first partial derivatives vanish at infinity.
50
1 Linear Partial Differential Equations
We apply the two-dimensional Fourier transform defined by (1.8.3) to the system (1.8.26), (1.8.27ab), which becomes d2 U + c2 κ2 U = 0, dt2
κ2 = k2 + 2 , dU = F (k, ). dt t=0
U (k, , 0) = 0,
The solution of this transformed system is U (k, , t) = F (k, )
sin(cκt) . cκ
The inverse Fourier transform gives the formal solution ∞ ∞ sin(cκt) 1 κ) dκ κ κ · r) F (κ exp(iκ u(x, y, t) = 2πc −∞ −∞ κ ∞ ∞ κ) κ·r 1 F (κ = exp iκ + ct 4iπc −∞ −∞ κ κ
κ·r κ. − ct dκ − exp iκ κ
(1.8.28)
(1.8.29)
(1.8.30)
The form of this solution reveals some interesting features of the wave equation. The exponential terms exp{iκ( κκ·r ± ct)} involved in the integral solution (1.8.30) represent plane wave solutions of the wave equation (1.8.26). Thus, the solutions remain constant on the planes κ · r = const. that move parallel to themselves with velocity c. Evidently, solution (1.8.30) represents a superposition of the plane wave solutions traveling in all possible directions. Similarly, the solution of the Cauchy problem for the three-dimensional wave equation utt = c2 (uxx + uyy + uzz ),
u(x, y, z, 0) = 0,
−∞ < x, y, z < ∞, t > 0,
ut (x, y, z, 0) = f (x, y, z),
(1.8.31)
−∞ < x, y, z < ∞, (1.8.32ab)
is given by κ) κ·r F (κ exp iκ + ct u(r, t) = 3 κ κ 2ic(2π) 2 −∞
κ·r κ, − exp iκ − ct dκ κ 1
∞
(1.8.33)
where r = (x, y, z) and κ = (k, , m). κ · r ± κct)} represent the plane traveling For any given κ , the terms exp{i(κ wave solution of the wave equation (1.8.31), since they remain constant on the planes
1.9 Laplace Transforms and Initial Boundary-Value Problems
51
κ at constant κ ·r = ±cκt, which move perpendicular to themselves in the direction ±κ speed c. Since the integral solution (1.8.33) is the weighted sum of these plane waves at different wavenumbers κ , the solution represents a continuous spectrum of plane waves propagating in all directions. It is somewhat easier to interpret solutions like (1.8.33) as t → ∞ with (r/t) fixed. This can be done using the method of stationary phase approximation. 3 κ) = (2π)− 2 , solution In particular, when f (x, y, z) = δ(x)δ(y)δ(z), so that F (κ (1.8.33) becomes ∞ 1 sin cκt κ · r) dκ κ. exp i(κ (1.8.34) u(r, t) = (2π)3 cκ −∞ In terms of the spherical polar coordinates (κ, θ, φ) where the polar axis (the zaxis) is taken along the r direction with κ · r = κr cos θ, we write (1.8.34) in the form π ∞ 2π sin cκt 2 1 dφ dθ exp(iκr cos θ) u(r, t) = · κ sin θ dκ (2π)3 0 cκ 0 0 ∞ 1 = sin(cκt) sin(κr) dκ 2π 2 cr 0 ∞ iκ(ct−r) 1 − eiκ(ct+r) dκ, = Re e 2 8π cr −∞ or u(r, t) =
1 δ(ct − r) − δ(ct + r) . 4πcr
For t > 0, ct + r > 0, so that δ(ct + r) = 0 and, hence, the solution is 1 1 r u(r, t) = . δ(ct − r) = δ t− 4πcr 4πrc2 c
(1.8.35)
(1.8.36)
1.9 Laplace Transforms and Initial Boundary-Value Problems If u(x, t) is any function defined in a ≤ x ≤ b and t > 0, then its Laplace transform with respect to t is denoted by L{u(x, t)} = u ¯(x, s) and is defined by ∞ L u(x, t) = u ¯(x, s) = e−st u(x, t) dt, Re s > 0, (1.9.1) 0
where s is called the transform variable, which is a complex number. Under certain broad conditions on u(x, t), its transform u ¯(x, s) is an analytic function of s in the half plane Re s > c. u(x, s)} = u(x, t) and deThe inverse Laplace transform is denoted by L−1 {¯ fined by the complex integral
52
1 Linear Partial Differential Equations
1 ¯(x, s) = u(x, t) = L−1 u 2πi
c+i∞
est u ¯(x, s) ds,
c > 0.
(1.9.2)
c−i∞
This integral is evaluated by using the Cauchy residue theorem for analytic functions. For more information about the basic properties of the Laplace transforms including the convolution, see Section B-4, pp. 758–766. If L{u(x, t)} = u ¯(x, s), then (see Debnath 1995) ∂u L = s¯ u(x, s) − u(x, 0), (1.9.3) ∂t 2 n ∂ u n = s L u ¯ (x, s) − sn−r u(r−1) (x, 0), (1.9.4) ∂tn r=1 and so on. Similarly, it is easy to show that d¯ u ∂u = , L ∂x dx
∂ 2u L ∂x2
=
¯ ∂2u . ∂x2
The following examples are useful for applications: ¯ L f (t − a)H(t − a) = exp(−sa)f (s), a > 0, δ(t − a) L = exp(−as), a > 0, √ a 1 L erfc √ = exp −a s , a ≥ 0, s √ 2 t √ a L exp(at)erf at = √ , a > 0. s(s − a)
(1.9.5ab)
(1.9.6) (1.9.7) (1.9.8) (1.9.9)
The Laplace transforms are also very useful in finding solutions of initial-value problems described by linear partial differential equations. The following examples of applications illustrate the method of Laplace transforms. Example 1.9.1 (The Inhomogeneous Cauchy Problem for the Wave Equation). We use the joint Laplace and Fourier transform method to solve the inhomogeneous Cauchy problem utt − c2 uxx = q(x, t), u(x, 0) = f (x),
x ∈ R (i.e., − ∞ < x < ∞), t > 0, ut (x, 0) = g(x)
for all x ∈ R,
(1.9.10) (1.9.11ab)
where q(x, t) is a given function representing a source term. We define the joint Laplace and Fourier transform of u(x, t) by ∞ ∞ 1 U (k, s) = L F u(x, t) = √ e−ikx dx e−st u(x, t) dt. 2π −∞ 0 (1.9.12) Application of the joint transform leads to the solution of the transformed problem in the form
1.9 Laplace Transforms and Initial Boundary-Value Problems
U (k, s) =
sF (k) + G(k) + Q(k, s) . (s2 + c2 k 2 )
53
(1.9.13)
The inverse Laplace transform of (1.9.13) gives ck 1 1 · Q(k, s) G(k) sin(ckt) + L−1 2 ck ck s + c2 k 2 t G(k) 1 sin ck(t − τ )Q(k, τ ) dτ. = F (k) cos(ckt) + sin(ckt) + ck ck 0 (1.9.14)
U (k, t) = F (k) cos(ckt) +
The inverse Fourier transform leads to the solution ∞ ickt 1 u(x, t) = √ + e−ickt eikx F (k) dk e 2 2π −∞ ∞ ickt G(k) 1 − e−ickt eikx · + √ dk e ick 2 2π −∞ t ∞ Q(k, τ ) ick(t−τ ) 1 1 e dτ + e−ick(t−τ ) eikx dk +√ · ik 2π 2c 0 −∞ x+ct 1 1 g(ξ) dξ = f (x + ct) + f (x − ct) + 2 2c x−ct x+c(t−τ ) ∞ 1 1 t dτ √ Q(k, τ ) dk eikξ dξ + 2c 0 2π −∞ x−c(t−τ ) x+ct 1 1 g(ξ) dξ = f (x − ct) + f (x + ct) + 2 2c x−ct x+c(t−τ ) 1 t dτ q(ξ, τ ) dξ. (1.9.15) + 2c 0 x−c(t−τ ) In the case of the homogeneous Cauchy problem, q(x, t) ≡ 0, the solution of (1.9.15) reduces to the famous d’Alembert solution (1.7.21). Example 1.9.2 (The Heat Conduction Equation in a Semi-Infinite Medium and Fractional Derivatives). Solve the equation ut = κuxx ,
x > 0, t > 0,
(1.9.16)
with the initial and boundary conditions u(x, 0) = 0,
x > 0,
u(0, t) = f (t), u(x, t) → 0,
t > 0,
as x → ∞, t > 0.
(1.9.17) (1.9.18) (1.9.19)
54
1 Linear Partial Differential Equations
Application of the Laplace transform with respect to t to (1.9.16) gives s ¯ d2 u − u ¯ = 0. 2 dx κ
(1.9.20)
The general solution of this equation is s s u ¯(x, s) = A exp −x + B exp x , κ κ where A and B are integrating constants. For bounded solutions, B ≡ 0, and using u ¯(0, s) = f¯(s), we obtain the solution s ¯ . (1.9.21) u ¯(x, s) = f (s) exp −x κ The Laplace inversion theorem gives the solution t 3 x x2 dτ, u(x, t) = √ f (t − τ )τ − 2 exp − 4κτ 2 πκ 0 which is, by setting λ =
√x 2 κτ
2 = √ π
(1.9.22)
x or dλ = − 4√ τ − 2 dτ , κ 3
∞ x √ 2 κt
2 x2 e−λ dλ. f t− 2 4κλ
This is the formal solution of the heat conduction problem. In particular, if f (t) = T0 = const., the solution (1.9.23) becomes x 2T0 ∞ −λ2 . e dλ = T0 erfc √ u(x, t) = √ π √x 2 κt 2 κt
(1.9.23)
(1.9.24)
Clearly, the temperature distribution tends asymptotically to the constant value T0 , as t → ∞. We consider another physical problem that is concerned with determining the temperature distribution of a semi-infinite solid when the rate of flow of heat is prescribed at the end x = 0. Thus, the problem is to solve diffusion equation (1.9.16) subject to conditions (1.9.17), (1.9.19), and ∂u = g(t) at x = 0, t > 0, (1.9.25) −k ∂x where k is a constant called thermal conductivity. Application of the Laplace transform gives the solution of the transformed problem s 1 κ . (1.9.26) g(s) exp −x u ¯(x, s) = k s κ
1.9 Laplace Transforms and Initial Boundary-Value Problems
The inverse Laplace transform yields the solution x2 1 κ t − 12 dτ, g(t − τ )τ exp − u(x, t) = k π 0 4κτ which, by the change of variable λ = x = √ k π
∞ x √ 2 κt
55
(1.9.27)
√x , 2 κτ
2 x2 λ−2 e−λ dλ. g t− 2 4κλ
(1.9.28)
In particular, if g(t) = T0 = const., this solution becomes 2 T0 x ∞ u(x, t) = √ λ−2 e−λ dλ. x k π √ 2k
κt
Integrating this result by parts gives
κt x x2 T0 . 2 exp − − x erfc √ u(x, t) = k π 4κt 2 κt
(1.9.29)
Alternatively, the heat conduction problem (1.9.16)–(1.9.19) can be solved by using fractional derivatives (Debnath 1995). We recall (1.9.21) and rewrite it as ∂u ¯ s =− u ¯. (1.9.30) ∂x κ This can be expressed in terms of a fractional derivative of order
1 2
as
1 √ 1 ∂u 1 = − √ L−1 s¯ u(x, s) = − √ Dt2 u(x, t). ∂x κ κ0
(1.9.31)
Thus, the heat flux is expressed in terms of the fractional derivative. In particular, when u(0, t) = const. = T0 , then the heat flux at the surface is given by 1 ∂u k kT0 −k = √ Dt2 T0 = √ . (1.9.32) ∂x x=0 κ πκt Example 1.9.3 (Diffusion Equation in a Finite Medium). Solve the diffusion equation ut = κuxx ,
0 < x < a, t > 0,
(1.9.33)
with the initial and boundary conditions u(x, 0) = 0,
0 < x < a,
u(0, t) = U, t > 0, ux (a, t) = 0, t > 0, where U is a constant.
(1.9.34) (1.9.35) (1.9.36)
56
1 Linear Partial Differential Equations
We introduce the Laplace transform of u(x, t) with respect to t to obtain ¯ s d2 u ¯ = 0, 0 < x < a, − u 2 dx κ d¯ u U u ¯(0, s) = , = 0. s dx x=a
(1.9.37) (1.9.38ab)
The general solution of (1.9.37) is s s + B sinh x , u ¯(x, s) = A cosh x κ κ
(1.9.39)
where A and B are constants of integration. Using (1.9.38ab), we obtain the values of A and B, so that the solution (1.9.39) becomes U cosh[(a − x) κs ] s . (1.9.40) · u ¯(x, s) = s cosh(a κ ) The inverse Laplace transform gives the solution cosh(a − x) κs −1 u(x, t) = U L . s cosh a κs
(1.9.41)
The inversion can be carried out by the Cauchy residue theorem to obtain the solution ∞ (2n − 1)(a − x)π 4 (−1)n cos u(x, t) = U 1 + π n=1 (2n − 1) 2a 2 π κt . (1.9.42) × exp −(2n − 1)2 2a By expanding the cosine term, this becomes ∞ 1 4 2n − 1 u(x, t) = U 1 − πx sin π n=1 (2n − 1) 2a 2 π 2 κt . × exp −(2n − 1) 2a
(1.9.43)
This result can be obtained by solving the problem by the method of separation of variables. Example 1.9.4 (Diffusion in a Finite Medium). Solve the one-dimensional diffusion equation in a finite medium 0 < z < a, where the concentration function C(z, t) satisfies the equation Ct = κCzz ,
0 < z < a, t > 0,
(1.9.44)
1.9 Laplace Transforms and Initial Boundary-Value Problems
57
and the initial and boundary data C(z, 0) = 0 for 0 < z < a, C(z, t) = C0 for z = a, t > 0, ∂C = 0 for z = 0, t > 0, ∂z
(1.9.45) (1.9.46) (1.9.47)
where C0 is a constant. Application of the Laplace transform of C(z, t) with respect to t gives s d2 C − C = 0, 0 < z < a, dz 2 κ dC C0 , C(a, s) = = 0. s dz z=0 The solution of this differential equation system is C0 cosh(z κs ) C(z, s) = , s cosh(a κs ) which, by writing α =
(1.9.48)
s
κ,
C0 (eαz + e−αz ) s (eαa + e−αa ) ∞ C0 = exp −α(a − z) + exp −α(a + z) (−1)n exp(−2nαa) s n=0 ∞ C0 (−1)n exp −α (2n + 1)a − z = s n=0 ∞ n + . (1.9.49) (−1) exp −α (2n + 1)a + z =
n=0
Using the result (1.9.8), we obtain the final solution ∞ (2n + 1)a − z √ C(z, t) = C0 (−1)n erfc 2 κt n=0 (2n + 1)a + z √ . + erfc 2 κt
(1.9.50)
This solution represents an infinite series of complementary error functions. The successive terms of this series are, in fact, the concentrations at depth a − z, a + z, 3a − z, 3a + z, . . . in the medium. The series converges rapidly for all except large values of ( aκt2 ).
58
1 Linear Partial Differential Equations
Example 1.9.5 (The Wave Equation for the Transverse Vibration of a Semi-Infinite String). Find the displacement of a semi-infinite string, which is initially at rest in its equilibrium position. At time t = 0, the end x = 0 is constrained to move so that the displacement is u(0, t) = Af (t) for t ≥ 0, where A is a constant. The problem is to solve the one-dimensional wave equation utt = c2 uxx ,
0 ≤ x < ∞, t > 0,
(1.9.51)
with the boundary and initial conditions u(x, t) = Af (t) at x = 0, t ≥ 0, u(x, t) → 0 as x → ∞, t ≥ 0, ∂u at t = 0 for 0 < x < ∞. u(x, t) = 0 = ∂t
(1.9.52) (1.9.53) (1.9.54ab)
Application of the Laplace transform of u(x, t) with respect to t gives ¯ s2 d2 u − 2u ¯ = 0, for 0 ≤ x < ∞, dx2 c u ¯(x, s) = Af¯(s) at x = 0, u ¯(x, s) → 0 as x → ∞. The solution of this differential equation system is xs ¯ u ¯(x, s) = Af (s) exp − . c
(1.9.55)
Inversion gives the solution x x H t− . u(x, t) = Af t − c c
(1.9.56)
In other words, the solution is u(x, t) =
Af (t − xc )
if t > xc ,
0,
if t < xc .
(1.9.57)
This solution represents a wave propagating at a velocity c with the characteristic x = ct. Example 1.9.6 (The Cauchy–Poisson Wave Problem in Fluid Dynamics). We consider the two-dimensional Cauchy–Poisson problem (Debnath 1994) for an inviscid liquid of infinite depth with a horizontal free surface. We assume that the liquid has constant density ρ and negligible surface tension. Waves are generated on the free
1.9 Laplace Transforms and Initial Boundary-Value Problems
59
surface of water initially at rest for time t < 0 by the prescribed free surface displacement at t = 0. In terms of the velocity potential φ(x, z, t) and the free surface elevation η(x, t), the linearized surface wave motion in Cartesian coordinates (x, y, z) is governed by the following equation and free surface and boundary conditions: −∞ < z ≤ 0, x ∈ R, t < 0,
∇2 φ = φxx + φzz = 0, φz − ηt = 0, φt + gη = 0 φz → 0
(1.9.58)
on z = 0, t > 0, as z → −∞.
(1.9.59ab) (1.9.60)
The initial conditions are φ(x, 0, 0) = 0 and
η(x, 0) = η0 (x),
(1.9.61)
where η0 (x) is a given initial elevation with compact support. We introduce the Laplace transform with respect to t and the Fourier transform with respect to x defined by ∞ ∞ ˜ ˜ s) = √1 φ(k, z, s), η(k, e−ikx dx e−st [φ, η] dt. (1.9.62) 2π −∞ 0 Application of the joint transform method to the above system gives ˜ = 0, ˜ − k2φ φ zz
−∞ < z ≤ 0, ⎫ ˜ = sη˜ − η˜ (k), ⎬ φ 0 on z = 0, ⎭ ˜ + g η˜ = 0 sφ ˜ →0 φ z
as z → −∞,
(1.9.63) (1.9.64ab)
(1.9.65)
where η˜0 (k) = F η0 (x) . The bounded solution of equation (1.9.63) is ˜ s) = A exp|k|z , φ(k,
(1.9.66)
where A = A(s) is an arbitrary function of s. Substituting (1.9.66) into (1.9.64ab) and eliminating η˜ from the resulting equa˜ and η˜ are tions gives A. Hence, the solutions for φ
˜ g η˜0 exp(|k|z) s˜ η0 ˜ , (1.9.67) φ, η = − , 2 s2 + ω 2 s + ω2
60
1 Linear Partial Differential Equations
where the dispersion relation is ω 2 = g|k|.
(1.9.68)
The inverse Laplace and Fourier transforms give the solutions ∞ sin ωt g exp ikx + |k|z η˜0 (k) dk, φ(x, z, t) = − √ 2π −∞ ω ∞ 1 η(x, t) = √ η˜0 (k) cos ωteikx dk, 2π −∞ ∞ 1 = √ η˜0 (k) ei(kx−ωt) + ei(kx+ωt) dk, 2π 0
(1.9.69)
(1.9.70)
in which η˜0 (−k) = η˜0 (k) is assumed. Physically, the first and second integrals of (1.9.70) represent waves traveling in the positive and negative directions of x, respectively, with phase velocity ωk . These integrals describe superposition of all such waves over the wavenumber spectrum 0 < k < ∞. For the classical Cauchy–Poisson wave √ problem, η0 (x) = aδ(x) where δ(x) is the Dirac delta function, so that η˜0 (k) = a/ 2π. Thus, solution (1.9.70) becomes ∞ i(kx−ωt) a e (1.9.71) + ei(kx+ωt) dk. η(x, t) = 2π 0 The wave integrals (1.9.69) and (1.9.70) represent the exact solution for the velocity potential φ and the free surface elevation η for all x and t > 0. However, they do not lend any physical interpretations. In general, the exact evaluation of these integrals is almost a formidable task. So it is necessary to resort to asymptotic methods. It would be sufficient for the determination of the principal features of the wave motions to investigate (1.9.70) or (1.9.71) asymptotically for large time t and large distance x with (x, t) held fixed. The asymptotic solution for this kind of problem is available in many standard books; for example, see Debnath (1994, p. 85). We state the stationary phase approximation of a typical wave integral, for t → ∞, b f (k) exp itW (k) dk (1.9.72) η(x, t) = a
∼ f (k1 ) where W (k) = equation
kx t
2π t|W (k1 )|
12
π exp i tW (k1 ) + sgnW (k1 ) , 4
(1.9.73)
− ω(k), x > 0, and k = k1 is a stationary point that satisfies the W (k1 ) =
x − ω (k1 ) = 0, t
a < k1 < b.
(1.9.74)
Application of (1.9.73) to (1.9.70) shows that only the first integral in (1.9.70) has a stationary point for x > 0. Hence, the stationary phase approximation gives the asymptotic solution, as t → ∞, x > 0,
1.10 Hankel Transforms and Initial Boundary-Value Problems
1 η(x, t) ∼ t|ω (k1 )|
12
61
iπ η˜0 (k1 ) exp i k1 x − tω(k1 ) + sgn −ω (k1 ) , 4 (1.9.75)
where k1 = (gt2 /4x2 ) is the root of the equation ω (k) = xt . On the other hand, when x < 0, only the second integral of (1.9.70) has a stationary point k1 = (gt2 /4x2 ), and hence, the same result (1.9.73) can be used to obtain the asymptotic solution for t → ∞ and x < 0 as
1 η(x, t) ∼ t|ω (k1 )|
12
iπ η˜0 (k1 ) exp i tω(k1 ) − k1 |x| + sgnω (k1 ) . 4 (1.9.76)
In particular, for the classical Cauchy–Poisson solution (1.9.71), the asymptotic representation for η(x, t) follows from (1.9.76) in the form 2 √ g at gt η(x, t) ∼ √ cos (1.9.77) , gt2 4x, 3/2 4x 2 2π x and gives a similar result for η(x, t), when x < 0 and t → ∞.
1.10 Hankel Transforms and Initial Boundary-Value Problems The Hankel transform of a function f (r) is defined formally by ∞ ˜ rJn (κr)f (r) dr, Hn f (r) = fn (κ) =
(1.10.1)
0
where Jn (κr) is the Bessel function of order n and assuming the integral on the right-hand side is convergent. The inverse Hankel transform is defined by ∞ Hn−1 f˜n (κ) = f (r) = κJn (κr)f˜n (κ) dκ, (1.10.2) 0
provided that the integral exists. Integrals (1.10.1) and (1.10.2) exist for certain large classes of functions that usually occur in physical applications. In particular, the Hankel transforms of order zero (n = 0) and of order one (n = 1) are useful for solving initial-value and boundaryvalue problems involving Laplace’s or Helmholtz’s equations in an axisymmetric cylindrical geometry. Example 1.10.1. (a) f˜(κ) = H0 { 1r exp(−ar)} = (b) f˜(κ) = H0 { δ(r) } = 1, r
∞ 0
exp(−ar)J0 (κr) dr =
√ 1 , κ2 +a2
62
1 Linear Partial Differential Equations
a (c) f˜(κ) = H0 {H(a − r)} = 0 rJ0 (κr) dr = κa J1 (aκ), ∞ (d) f˜(κ) = H1 {e−ar } = 0 r exp(−ar)J1 (κr) dr = 2 κ 2 3 , (a +κ ) 2 ∞ −ar 1 (e) f˜(κ) = H1 { e r } = 0 e−ar J1 (κr) dr = κ1 [1 − a(κ2 + a2 )− 2 ]. It can be shown (Debnath 1995) that n2 2 ∇ − 2 f (r) = −κ2 f˜n (κ), Hn r
(1.10.3)
where df d2 f 1 d 1 df r = 2 + , ∇ f = r dr dr dr r dr 2
(1.10.4)
and rf (r) and rf (r) vanish, as r → 0 and r → ∞. In particular, when n = 0 and n = 1, (1.10.3) reduces to special results which are very useful for applications. Example 1.10.2. Obtain the solution of the boundary-value problem 1 urr + ur + uzz = 0, 0 ≤ r < ∞, z ≥ 0, r u(r, 0) = u0 for 0 ≤ r ≤ a, u0 is a constant, u(r, z) → 0 as z → ∞.
(1.10.5) (1.10.6) (1.10.7)
Application of the zero-order Hankel transform with respect to r to the problem gives d2 u ˜ − κ2 u ˜ = 0, dz 2 u ˜(κ, 0) = u0
a
rJ0 (κr) dr = 0
au0 J1 (aκ). κ
Thus, the solution of the transformed problem is u ˜(κ, z) =
au0 J1 (aκ) exp(−κz). κ
Using the inverse Hankel transform gives the formal solution ∞ u(r, z) = au0 J0 (rκ)J1 (aκ) exp(−κz) dκ.
(1.10.8)
0
Example 1.10.3 (Axisymmetric Wave Equation). Find the solution of the free vibration of a large circular membrane governed by the initial-value problem 1 2 (1.10.9) c urr + ur = utt , 0 ≤ r < ∞, t > 0, r
1.10 Hankel Transforms and Initial Boundary-Value Problems
u(r, 0) = f (r),
ut (r, 0) = g(r), 0 ≤ r ≤ ∞,
63
(1.10.10)
where c2 = (T /ρ) = const., T is the tension in the membrane, ρ is the surface density of the membrane, and f (r) and g(r) are arbitrary functions. Application of the zero-order Hankel transform u ˜(κ, t) of the displacement function u(r, t) to (1.10.9), (1.10.10) gives the solution u ˜(κ, t) = f˜(κ) cos(cκt) +
g˜(κ) sin(cκt). cκ
(1.10.11)
The inverse Hankel transform leads to the solution ∞ 1 ∞ ˜ κf (κ) cos(cκt)J0 (κr) dκ + g˜(κ) sin(cκt)J0 (κr) dκ. u(r, t) = c 0 0 (1.10.12) In particular, we consider the initial data − 1 u(r, 0) = f (r) = Aa r2 + a2 2 ,
ut (r, 0) = g(r) = 0, (1.10.13ab)
so that g˜(κ) ≡ 0 and ∞ − 1 Aa −aκ ˜ f (κ) = Aa e r a2 + r2 2 J0 (κr) dr = , κ 0
by Example 1.10.1(a).
Thus, the formal solution (1.10.12) becomes ∞ e−aκ J0 (κr) cos(cκt) dκ u(r, t) = Aa 0 ∞ = Aa Re exp −κ(a + ict) J0 (κr) dκ
0
= Aa Re r2 + (a + ict)2
− 12
,
by Example 1.10.1(a).
(1.10.14)
Example 1.10.4 (Steady Temperature Distribution in a Semi-Infinite Solid with a Steady Heat Source). Find the solution of the Laplace equation for the steady temperature distribution u(r, z) with a steady, symmetric heat source Q0 q(r): 1 urr + ur + uzz = −Q0 q(r), 0 ≤ r < ∞, 0 < z < ∞, r u(r, 0) = 0, 0 ≤ r < ∞,
(1.10.15) (1.10.16)
where Q0 is constant. This boundary condition represents zero temperature at the boundary z = 0. Application of the zero-order Hankel transform to (1.10.15), (1.10.16) gives ˜ d2 u − κ2 u ˜ = −Q0 q˜(κ), dz 2
u ˜(κ, 0) = 0.
64
1 Linear Partial Differential Equations
The bounded general solution of this system is u ˜(κ, z) = A exp(−κz) +
Q0 q˜(κ), κ2
where A is a constant to be determined from the transformed boundary condition. In this present case, A=−
Q0 q˜(κ). κ2
Thus, the formal solution is given by u ˜(κ, z) =
Q0 q˜(κ) 1 − e−κz . κ2
The inverse Hankel transform yields the exact integral solution ∞ q˜(κ) u(r, z) = Q0 1 − e−κz J0 (κr) dκ. κ 0
(1.10.17)
(1.10.18)
Example 1.10.5 (Axisymmetric Diffusion Equation). Find the solution of the axisymmetric diffusion equation 1 ut = κ urr + ur , 0 ≤ r < ∞, t > 0, (1.10.19) r where κ(> 0) is a diffusivity constant, with the initial condition u(r, 0) = f (r),
for 0 < r < ∞.
(1.10.20)
We apply the zero-order Hankel transform to obtain d˜ u + k2 κ˜ u = 0, dt
u ˜(k, 0) = f˜(k),
where k is the Hankel transform variable. The solution of this transformed system is (1.10.21) u ˜(k, t) = f˜(k) exp −κk2 t . Application of the inverse Hankel transform gives ∞ 2 u(r, t) = k f˜(k)J0 (kr)e−κk t dk
0 ∞ ∞ 2 k lJ0 (kl)f (l) dl e−κk t J0 (kr) dk, = 0
0
which, interchanging the order of integration, ∞ ∞ lf (l) dl kJ0 (kl)J0 (kr) exp −κk2 t dk. = 0
0
(1.10.22)
1.10 Hankel Transforms and Initial Boundary-Value Problems
65
Using a standard table of integrals involving Bessel functions, we state that
∞ 1 rl (r2 + l2 ) 2 kJ0 (kl)J0 (kr) exp −κk t dk = I0 , exp − 2κt 4κt 2κt 0 (1.10.23) where I0 (x) is the modified Bessel function and I0 (0) = 1. In particular, when l = 0, J0 (0) = 1, and integral (1.10.23) becomes ∞ 2 r2 1 . (1.10.24) exp − kJ0 (kr) exp −k κt dk = 2κt 4κt 0 We next use (1.10.23) to rewrite the solution (1.10.22) as
∞ 1 (r 2 + l2 ) rl u(r, t) = exp − dl. lf (l)I0 2κt 0 2κt 4κt
(1.10.25)
We now assume that f (r) represents a heat source concentrated in a circle of radius a and allow a → 0, so that the heat source is concentrated at r = 0 and a lim 2π rf (r) dr = 1. a→0
0
Or equivalently, f (r) =
1 δ(r) , 2π r
where δ(r) is the Dirac delta function. Thus, the final solution due to the concentrated heat source at r = 0 is 2
∞ r + l2 rl 1 δ(l)I0 exp − dl u(r, t) = 4πκt 0 2κt 4κt 1 r2 = . (1.10.26) exp − 4πκt 4κt Example 1.10.6 (Axisymmetric Acoustic Radiation Problem). Obtain the solution of the wave equation 1 (1.10.27) c2 urr + ur + uzz = utt , 0 ≤ r < ∞, t > 0, r uz = F (r, t) on z = 0, (1.10.28) where F (r, t) is a given function and c is a constant. We also assume that the solution is bounded and behaves as outgoing spherical waves. We seek a steady-state solution for the acoustic radiation potential u = eiωt × φ(r, z) with F (r, t) = eiωt f (r), so that φ satisfies the Helmholtz equation
66
1 Linear Partial Differential Equations
1 ω2 φrr + φr + φzz + 2 φ = 0, r c
0 ≤ r < ∞, z ≥ 0,
(1.10.29)
with the boundary condition φz = f (r)
on z = 0,
(1.10.30)
where f (r) is a given function of r. ˜ z) to (1.10.29), Application of the Hankel transform H0 {φ(r, z)} = φ(k, (1.10.30) gives ˜ z > 0, φ˜zz = κ2 φ, ˜ ˜ φz = f (k), on z = 0, where κ=
k2 −
ω2 c2
12 .
The solution of this differential equation system is ˜ z) = − 1 f˜(k) exp(−κz), φ(k, κ
(1.10.31)
where κ is real and positive for k > ω/c and purely imaginary for k < ω/c. The inverse Hankel transform yields the formal solution ∞ k˜ (1.10.32) φ(r, z) = − f (k)J0 (kr) exp(−κz) dk. κ 0 Since the exact evaluation of this integral is difficult for an arbitrary f˜(k), we choose a simple form of f (r) as f (r) = AH(a − r),
(1.10.33)
where A is a constant, and hence, f˜(k) = Aa k J1 (ak). Thus, the solution (1.10.32) takes the form ∞ 1 φ(r, z) = −Aa J1 (ak)J0 (kr) exp(−κz) dk. κ 0
(1.10.34)
For an asymptotic evaluation of this integral, it is convenient to express (1.10.34) in terms of the distance R from the z-axis, so that R2 = (r2 + z 2 ) and z = R cos θ. We use the asymptotic result for the Bessel function in the form J0 (kr) ∼
2 πkr
12
π cos kr − 4
as r → ∞,
(1.10.35)
where r = R sin θ. Consequently, (1.10.34) combined with u = exp(iωt)φ becomes
1.10 Hankel Transforms and Initial Boundary-Value Problems
67
√ Aa 2eiωt ∞ 1 π √ J1 (ak) cos kR sin θ − exp(−κz) dk. u ∼ −√ 4 πR sin θ 0 κ k This integral can be evaluated asymptotically for R → ∞, by using the stationary phase approximation formula to obtain the final result
Aac ωR u∼− , (1.10.36) J1 (ak1 ) exp i ωt − ωR sin θ c where k1 = ω/c sin θ is the stationary point. Physically, this solution represents outgoing spherical waves with constant velocity c and decaying amplitude, as R → ∞. Example 1.10.7 (Axisymmetric Biharmonic Equation). We solve the axisymmetric boundary-value problem ∇4 u(r, z) = 0,
0 ≤ r < ∞, z > 0
(1.10.37)
u(r, 0) = f (r), 0 ≤ r < ∞, ∂u = 0 on z = 0, 0 ≤ r < ∞, ∂z u(r, z) → 0 as r → ∞,
(1.10.38)
with the boundary data
where the axisymmetric biharmonic operator is 2 2 2 ∂ ∂2 ∂2 ∂ 1 ∂ 1 ∂ 4 2 . + + ∇ =∇ ∇ = + + ∂r2 r ∂r ∂z 2 ∂r2 r ∂r ∂z 2
(1.10.39) (1.10.40)
(1.10.41)
The use of the Hankel transform H0 {u(r, z)} = u ˜(k, z) for this problem gives
d2 − k2 dz 2
2 u ˜(k, z) = 0,
z > 0,
u ˜(k, 0) = f˜(k),
d˜ u = 0 on z = 0. dz
(1.10.42) (1.10.43)
The bounded solution of equation (1.10.42) is u ˜(k, z) = (A + zB) exp(−kz),
(1.10.44)
where A and B are integration constants to be determined by (1.10.43) as A = f˜(k) and B = k f˜(k). Thus, solution (1.10.44) becomes u ˜(k, z) = (1 + kz)f˜(k) exp(−kz). The inverse Hankel transform gives the formal solution ∞ k(1 + kz)f˜(k)J0 (kr) exp(−kz) dk. u(r, z) = 0
(1.10.45)
(1.10.46)
68
1 Linear Partial Differential Equations
Example 1.10.8 (The Axisymmetric Cauchy–Poisson Water Wave Problem). We consider the initial-value problem (Debnath 1994) for inviscid water of finite depth h with a free horizontal surface at z = 0 and the z-axis positive upward. We assume that the liquid has constant density ρ with no surface tension. The surface waves are generated in water which is initially at rest for t < 0 by the prescribed free surface elevation. In cylindrical polar coordinates (r, θ, z), the axisymmetric water wave equations for the velocity potential φ(r, z, t) and the free surface elevation η(r, t) are 1 ∇2 φ = φrr + φr + φzz = 0, r φz − ηt = 0, φt + gη = 0 φz = 0,
0 ≤ r < ∞, −h ≤ z ≤ 0, t > 0, on z = 0, t > 0,
on z = −h, t > 0,
(1.10.47) (1.10.48ab) (1.10.49)
where g is the constant gravitational acceleration. The initial conditions are φ(r, 0, 0) = 0 and
η(r, 0) = η0 (r),
for 0 ≤ r < ∞,
(1.10.50ab)
where η0 (r) is the prescribed free surface elevation. We apply the joint Laplace and the zero-order Hankel transform defined by ∞ ∞ ˜ z, s) = φ(k, e−st dt rJ0 (kr)φ(r, z, t) dr, (1.10.51) 0
0
to (1.10.47)–(1.10.49) so that these equations reduce to 2 d 2 ˜ − k φ = 0, −h ≤ z ≤ 0, dz 2 ⎫ ˜ dφ ⎬ ˜ = −˜ − s η η (k), 0 dz on z = 0, ⎭ ˜ + g η˜ = 0 sφ ˜ = 0 on z = −h, φ z
where η˜0 (k) is the zero-order Hankel transform of η0 (r). The solutions of this system are ˜ z, s) = − g η˜0 (k) cosh k(z + h) , φ(k, (s2 + ω 2 ) cosh kh η0 (k) ˜ z, s) = s˜ η(k, , (s2 + ω 2 )
(1.10.52) (1.10.53)
where ω 2 = gk tanh(kh)
(1.10.54)
1.10 Hankel Transforms and Initial Boundary-Value Problems
69
is the famous dispersion relation between frequency ω and wavenumber k for water waves in a liquid of depth h. Physically, this dispersion relation describes the interaction between the inertial and gravitational forces. Application of the inverse transforms gives the integral solutions ∞ sin ωt cosh k(z + h) kJ0 (kr)˜ η0 (k) φ(r, z, t) = −g dk, (1.10.55) ω cosh kh 0 ∞ kJ0 (kr)˜ η0 (k) cos ωt dk. (1.10.56) η(r, t) = 0
These wave integrals represent exact solutions for φ and η for all r and t, but the physical features of the wave motions cannot be described by them. In general, the exact evaluation of the integrals is almost a formidable task. In order to resolve this difficulty, it is necessary and useful to resort to asymptotic methods. It will be sufficient for the determination of the basic features of the wave motions to evaluate (1.10.55) or (1.10.56) asymptotically for a large time and distance with (r/t) held fixed. We now replace J0 (kr) by its asymptotic formula (1.10.35) for kr → ∞, so that (1.10.56) gives 12 ∞ √ 2 π cos ωt dk η(r, t) ∼ k η˜0 (k) cos kr − πr 4 0
∞√ 1 π = (2πr)− 2 Re dk. (1.10.57) k η˜0 (k) exp i ωt − kr + 4 0 Application of the stationary phase method to (1.10.57) yields the solution
12 k1 η˜0 (k1 ) cos tω(k1 ) − k1 r , (1.10.58) η(r, t) ∼ rt|ω (k1 )| where the stationary point k1 = (gt2 /4r2 ) is the root of the equation ω (k) =
r . t
(1.10.59)
For sufficiently deep water, kr → ∞, the dispersion relation becomes ω2 = gk.
(1.10.60)
The solution of the axisymmetric Cauchy–Poisson problem is obtained by using a prescribed initial displacement of unit volume concentrated at the origin, which a means that η0 (r) = (a/2πr)δ(r), so that η˜0 (k) = 2π . Thus, the asymptotic solution is obtained from (1.10.58) in the form 2 gt agt2 √ , gt2 4r. cos (1.10.61) η(r, t) ∼ 3 4r 4π 2r It is noted that solution (1.10.58) is no longer valid when ω (k1 ) = 0. This case can be handled by a modification of the asymptotic evaluation (see Debnath 1994, p. 91).
70
1 Linear Partial Differential Equations
1.11 Green’s Functions and Boundary-Value Problems Many physical problems are described by second-order nonhomogeneous differential equations with homogeneous boundary conditions or by second-order homogeneous equations with nonhomogeneous boundary conditions. Such problems can be solved by a powerful method based on a device known as Green’s functions. We consider a nonhomogeneous partial differential equation of the form Lx u(x) = f (x),
(1.11.1)
where x = (x, y, z) is a vector in three (or higher) dimensions, Lx is a linear partial differential operator in three or more independent variables with constant coefficients, and u(x) and f (x) are functions of three or more independent variables. The Green function G(x, ξ ) of this problem satisfies the equation Lx G(x, ξ ) = δ(x − ξ )
(1.11.2)
and represents the effect at the point x of the Dirac delta function source of the point ξ = (ξ, η, ζ). Multiplying (1.11.2) by f (ξξ ) and integrating over the volume V of the ξ space, so that dV = dξ dη dζ, we obtain Lx G(x, ξ )f (ξξ ) dξξ = δ(x − ξ )f (ξξ ) dξξ = f (x). (1.11.3) V
V
Interchanging the order of the operator Lx and integral sign in (1.11.3) gives
G(x, ξ )f (ξξ ) dξξ = f (x). (1.11.4) Lx V
A simple comparison of (1.11.4) with (1.11.1) leads to the solution of (1.11.1) in the form u(x) = G(x, ξ )f (ξξ ) dξξ . (1.11.5) V
Clearly, (1.11.5) is valid for any infinite number of components of x. Accordingly, the Green’s function method can be applied, in general, to any linear, constant coefficient, inhomogeneous partial differential equations in any number of independent variables. Another way to approach the problem is by looking for the inverse operator L−1 x . If it is possible to find L−1 , then the solution of (1.11.1) can be obtained as u(x) = x (f (x)). It turns out that, in many important cases, it is possible, and the inverse L−1 x operator can be expressed as an integral operator of the form ξ f (ξ ) = u(x) = L−1 G(x, ξ )f (ξξ ) dξξ . (1.11.6) x V
The kernel G(x, ξ ) is called the Green’s function which is, in fact, the characteristic of the operator Lx for any finite number of independent variables. The main goal of this section is to develop a general method of Green’s function from several examples of applications.
1.11 Green’s Functions and Boundary-Value Problems
71
Example 1.11.1 (Green’s Function for the One-Dimensional Diffusion Equation). We consider the inhomogeneous, one-dimensional, diffusion equation ut − κuxx = f (x)δ(t),
x ∈ R, t > 0,
(1.11.7)
with the boundary conditions u(x, 0) = 0 for x ∈ R
and
u(x, t) → 0
as |x| → ∞.
(1.11.8ab)
We take the Laplace transform with respect to t and the Fourier transform with respect to x to (1.11.7), (1.11.8ab), so that f˜(k) , (s + κk2 ) ˜ u ¯(k, t) → 0 as |k| → ∞.
˜ u ¯(k, s) =
(1.11.9) (1.11.10)
The inverse Laplace transform gives u ˜(k, t) = f˜(k) exp −κk2 t = f˜(k)˜ g (k),
(1.11.11)
where g˜(k) = exp(−κk 2 t), so that g(x) = F
−1
x2 1 2 . exp − exp −κk t = √ 4κt 2κt
(1.11.12)
Application of the inverse Fourier transform combined with Convolution Theorem 1.7.1 gives ∞ ∞ 1 1 ikx ˜ e f (k)˜ g (k) dκ = √ f˜(ξ)˜ g (x − ξ) dξ u(x, t) = √ 2π −∞ 2π −∞ ∞ (x − ξ)2 1 f˜(ξ) exp − dξ = √ 4κt 4κπt −∞ ∞ 1 f˜(ξ)G(x, t; ξ) dξ, (1.11.13) = √ 2π −∞ where the Green function G(x, t; ξ) is given by
(x − ξ)2 1 . exp − G(x, t; ξ) = √ 4κt 4πκt
(1.11.14)
Evidently, G(x, t) = G(x, t; 0) is an even function of x, and at any time t, the spatial distribution of G(x, √t) is Gaussian. The amplitude (or peak height) of G(x, √ t) decreases inversely with κt, whereas the width of the peak increases with κt. The evolution of√G(x, t) = u(x, t) has already been plotted against x for different values of τ = 2 κt in Figure 1.6.
72
1 Linear Partial Differential Equations
Example 1.11.2 (Green’s Function for the Two-Dimensional Diffusion Equation). We consider the two-dimensional diffusion equation ut − K∇2 u = f (x, y)δ(t),
−∞ < x, y < ∞, t > 0,
(1.11.15)
with the initial and boundary conditions u(x, y, 0) = 0, u(x, y, t) → 0
for all (x, y) ∈ R2 , as r = x2 + y 2 → ∞,
(1.11.16) (1.11.17)
where K is the diffusivity constant. Application of the Laplace transform and the double Fourier transform to the above system gives ˜ u ¯(κ, s) =
f˜(k, l) , (s + Kκ2 )
where κ = (k, l). The inverse Laplace transform gives g (k, l), u ˜(κ, t) = f˜(k, l) exp −Kκ2 t = f˜(k, l)˜
(1.11.18)
(1.11.19)
where g˜(k, l) = exp(−Kκ2 t), so that g(x, y) = F −1 exp −Kκ2 t =
(x2 + y 2 ) 1 . exp − 2Kt 4Kt
(1.11.20)
Finally, the convolution theorem of the Fourier transform gives the formal solution
∞ 1 (x − ξ)2 + (y − η)2 u(x, y, t) = dξ dη, f (ξ, η) exp − 4πKt 4Kt −∞ (1.11.21) ∞ ∞ f (ξξ )G(r, ξ ) dξξ , (1.11.22) = −∞
−∞
where r = (x, y) and ξ = (ξ, η) and Green’s function G(r, ξ ) is given by
1 |r − ξ |2 G(r, ξ ) = . (1.11.23) exp − (4πKt) 4Kt Similarly, we can construct Green’s function for the three-dimensional diffusion equation ut − K∇2 u = f (r)δ(t) with the initial and boundary data
for −∞ < x, y, z < ∞, t > 0,
(1.11.24)
1.11 Green’s Functions and Boundary-Value Problems
u(r, 0) = 0 for −∞ < x, y, z < ∞, 1 u(r, t) → 0 as r = x2 + y 2 + z 2 2 → ∞,
73
(1.11.25) (1.11.26)
where r = (x, y, z). Application of the Laplace transform of u(r, t) with respect to t and the threedimensional Fourier transform with respect to x, y, z gives the solution
∞ 1 |r − ξ |2 u(r, t) = dξξ , (1.11.27) f (ξξ ) exp − 3 4Kt (4πKt) 2 −∞ ∞ ∞ ∞ = f (ξξ )G(r, ξ ) dξξ , (1.11.28) −∞
−∞
−∞
where ξ = (ξ, η, ζ), and Green’s function is given by
|r − ξ |2 1 exp − G(r, ξ ) = . 3 4Kt (4πKt) 2
(1.11.29)
In fact, the same method of construction can be used to find the Green function of the n-dimensional diffusion equation ut − K∇2n u = f (r)δ(t), r ∈ Rn , t > 0, u(r, 0) = 0 for all r ∈ Rn , 1 u(r, t) → 0 as r = x21 + x22 + · · · + x2n 2 → ∞,
(1.11.30) (1.11.31) (1.11.32)
where r = (x1 , x2 , . . . , xn ) and ∇2n is the n-dimensional Laplacian given by ∇2n =
∂2 ∂2 ∂2 + + · · · + . ∂x21 ∂x22 ∂x2n
The solution of this problem is given by ∞ 1 u(r, t) = f (ξξ )G(r, ξ ) dξξ , n (4πKt) 2 −∞
(1.11.33)
(1.11.34)
where ξ = (ξ1 , ξ2 , . . . , ξn ) and the n-dimensional Green function G(r, ξ ) is given by
1 |r − ξ |2 G(r, ξ ) = . (1.11.35) exp − n 4Kt (4πKt) 2 Example 1.11.3 (The Three-Dimensional Poisson Equation). We show that the solution of the Poisson equation −∇2 u = f (r), where r = (x, y, z) is given by
(1.11.36)
74
1 Linear Partial Differential Equations
∞
∞
∞
G(r, ξ )f (ξξ ) dξξ ,
u(r) = −∞
−∞
(1.11.37)
−∞
where Green’s function G(r, ξ ) of the operator, −∇2 , is given by G(r, ξ ) =
1 1 . 4π |r − ξ |
(1.11.38)
To obtain the fundamental solution, we need to solve the equation −∇2 G(r − ξ ) = δ(x − ξ)δ(y − η)δ(z − ζ),
r = ξ .
(1.11.39)
Application of the three-dimensional Fourier transform, defined by (1.8.5), to (1.11.39) gives κ, ξ ) = κ2 G(κ
1 κ · ξ ), exp(−iκ (2π)3/2
κ, ξ ) = F{G(r, ξ )} and κ = (k, , m). where G(κ The inverse Fourier transform gives the formal solution ∞ ∞ ∞ dκ κ 1 κ · (r − ξ ) 2 exp iκ G(r, ξ ) = 3 (2π) −∞ −∞ −∞ κ ∞ ∞ ∞ κ 1 dκ κ · R) 2 , = exp(iκ (2π)3 −∞ −∞ −∞ κ
(1.11.40)
(1.11.41)
where R = r − ξ . We evaluate this integral using the spherical polar coordinates in the κ -space with the axis along the R-axis. In terms of spherical polar coordinates (κ, θ, φ), κ · R = κR cos θ where R = |r − ξ |. Thus, (1.11.41) becomes 2π π ∞ dκ 1 dφ dθ exp(κR cos θ)κ2 sin θ · 2 G(r, ξ ) = (2π)3 0 κ 0 0 ∞ 1 sin(κR) 1 1 = 2 dκ = = , (1.11.42) (2π)2 0 κR 4πR 4π|r − ξ | provided that R > 0. In electrodynamics, the fundamental solution (1.11.42) has a well-known interpretation. Physically, it represents the potential at point r generated by the unit point charge distribution at point ξ . This is what can be expected because δ(r − ξ ) is the charge density corresponding to a unit point charge at ξ . The solution of (1.11.36) is then given by ∞ ∞ 1 f (ξξ ) dξξ . (1.11.43) G(r, ξ )f (ξξ ) dξξ = u(r) = 4π −∞ −∞ |r − ξ | The integrand in (1.11.43) consists of the given charge distribution f (r) at r = ξ and Green’s function G(r, ξ ). Physically, G(r, ξ )f (ξξ ) represents the resulting potentials due to elementary point charges, and the total potential due to a given charge distribution f (r) is then obtained by the integral superposition of the resulting potentials. This is called the principle of superposition.
1.11 Green’s Functions and Boundary-Value Problems
75
Example 1.11.4 (The Two-Dimensional Helmholtz Equation). Find the fundamental solution of the two-dimensional Helmholtz equation −∇2 G + α2 G = δ(x − ξ)δ(y − η),
−∞ < x, y < ∞.
(1.11.44)
It is convenient to change variables x − ξ = x∗ , y − η = y∗ . Consequently, dropping the asterisks, (1.11.44) reduces to the form Gxx + Gyy − α2 G = −δ(x)δ(y).
(1.11.45)
κ) = F{G(x, y)} to (1.11.45) Application of the double Fourier transform G(κ gives the solution as 1 κ) = 1 G(κ , 2 2π (κ + α2 )
(1.11.46)
where κ = (k, ) and κ2 = k 2 + 2 . The inverse Fourier transform yields the solution ∞ ∞ −1 1 κ · x) κ2 + α2 exp(iκ dk d. G(x, y) = 4π 2 −∞ −∞
(1.11.47)
In terms of polar coordinates (x, y) = r(cos θ, sin θ), (k, ) = ρ(cos φ, sin φ), the integral solution (1.11.47) becomes ∞ 2π ρ dρ 1 G(x, y) = exp irρ cos(φ − θ) dφ, 4π 2 0 (ρ2 + α2 ) 0 which, replacing the second integral by 2πJ0 (rρ), ∞ ρJ0 (rρ) dρ 1 . = 2π 0 (ρ2 + α2 )
(1.11.48)
In terms of the original coordinates, the fundamental solution of (1.11.44) is given by G(r, ξ ) =
1 2π
∞ 0
1
ρJ0 [ρ{(x − ξ)2 + (y − η)2 } 2 ] dρ . (ρ2 + α2 )
Accordingly, the solution of the inhomogeneous equation 2 ∇ − α2 u = −f (x, y)
(1.11.49)
(1.11.50)
is given by
∞
∞
G(r, ξ )f (ξξ ) dξξ ,
u(x, y) = −∞
where G(r, ξ ) is given by (1.11.49).
−∞
(1.11.51)
76
1 Linear Partial Differential Equations
Since the integral solution (1.11.48) does not exist for α = 0, Green’s function for the two-dimensional Poisson equation (1.11.44) cannot be derived from (1.11.48). Instead, we differentiate (1.11.48) with respect to r to obtain ∞ ρJ0 (rρ) dρ 1 ∂G = , ∂r 2π 0 (ρ2 + α2 ) which is, for α = 0, ∂G 1 = ∂r 2π
∞
0
1 1 J0 (rρ) dρ = − . ρ 2πr
Integrating this result gives Green’s function G(r, θ) = −
1 log r. 2π
In terms of the original coordinates, Green’s function becomes G(r, ξ ) =
1 log (x − ξ)2 + (y − η)2 . 4π
(1.11.52)
This is Green’s function for the two-dimensional Poisson equation ∇2 = −f (x, y). Thus, the solution of the Poisson equation is ∞ ∞ G(r, ξ )f (ξξ ) dξξ , (1.11.53) u(x, y) = −∞
−∞
where G(r, ξ ) is given by (1.11.52). Example 1.11.5 (Green’s Function of the Three-Dimensional Helmholtz Equation). We consider the three-dimensional wave equation utt − c2 ∇2 u = q(r, t), (1.11.54) where q(r, t) is a source. If q(r, t) = q(r) exp(−iωt) represents a source oscillating with a single frequency ω, then, as expected, at least after an initial transient period, the entire motion reduces to a wave motion with the same frequency ω so that we can write u(r, t) = u(r) exp(−iωt). Consequently, the wave equation (1.11.54) reduces to the three-dimensional Helmholtz equation − ∇2 + k2 u(r) = f (r), (1.11.55) where k = ωc and f (r) = c−2 q(r). The function u(r) satisfies this equation on some domain D ⊂ R3 with boundary ∂D, and it also satisfies some prescribed boundary conditions. We also assume that u(r) satisfies the Sommerfeld radiation condition which simply states that the solution behaves like outgoing waves generated by the source. In the limit as ω → 0, so that k → 0 and f (r) can be interpreted as a heat source, equation (1.11.55) results in a three-dimensional Poisson equation. The
1.11 Green’s Functions and Boundary-Value Problems
77
solution u(r) would represent the steady temperature distribution in D due to the heat source f (r). However, in general, u(r) can be interpreted as a function of physical interest. We construct a Green function G(r, ξ ) for equation (1.11.55) so that G(r, ξ ) satisfies the equation (1.11.56) − ∇2 + k 2 G = δ(x)δ(y)δ(z). Using the spherical polar coordinates, the three-dimensional Laplacian can be expressed in terms of radial coordinate r only so that (1.11.56) assumes the form
∂G 1 ∂ δ(r) r2 + k2 G = − 2 , 0 < r < ∞, (1.11.57) r ∂r ∂r 4πr2 with the radiation condition lim r(Gr + ikG) = 0.
r→∞
For r > 0, the function G satisfies the homogeneous equation 1 ∂ 2 ∂G r + k 2 G = 0, r 2 ∂r ∂r
(1.11.58)
(1.11.59)
or equivalently, ∂ (rG) + k2 (rG) = 0. ∂r
(1.11.60)
This equation admits a solution of the form rG(r) = Aeikr + Be−ikr ,
(1.11.61)
eikr e−ikr +B , r r
(1.11.62)
or G(r) = A
where A and B are arbitrary constants. In order to satisfy the radiation condition, we need to set A = 0, and hence, the solution (1.11.62) becomes G(r) = B
e−ikr . r
To determine B, we use the spherical surface Sε of radius ε, so that ∂G B −ikr 1 lim dS = − lim e + ik dS = 1, ε→0 S ∂r ε→0 S r r ε ε
(1.11.63)
(1.11.64)
1 from which we find B = − 4π as ε → 0. Consequently, Green’s function takes the form
78
1 Linear Partial Differential Equations
G(r) = −
e−ikr . 4πr
(1.11.65)
Physically, this represents outgoing spherical waves radiating away from the source at the origin. With a point source at a point ξ , Green’s function is represented by G(r, ξ ) = −
exp{−ik|r − ξ |} , 4π|r − ξ |
(1.11.66)
where r and ξ are position vectors in R3 . Finally, when k = 0, this result reduces exactly to Green’s function for the threedimensional Poisson equation (1.11.36). Example 1.11.6 (One-Dimensional Wave Equation). We first consider the onedimensional inhomogeneous wave equation
1 (1.11.67) − uxx − 2 utt = c−2 q(x, t), x ∈ R, t > 0, c with the initial and boundary conditions u(x, 0) = 0,
ut (x, 0) = 0 for x ∈ R,
u(x, t) → 0
as |x| → ∞.
The Green function G(x, t) for this problem satisfies the equation
1 − Gxx − 2 Gtt = c−2 δ(x)δ(t) c
(1.11.68ab) (1.11.69)
(1.11.70)
and the same initial and boundary conditions (1.11.68ab), (1.11.69) satisfied by u(x, t). We apply the joint Laplace transform with respect to t and the Fourier transform with respect to x to equation (1.11.70), so that 1 c−2 ˜ G(k, s) = √ 2 , 2π (k 2 + sc2 )
(1.11.71)
where k and s represent the Fourier and Laplace transform variables, respectively. The inverse Fourier transform gives 1 1 s 1 1 −1 √ exp − |x| . (1.11.72) = G(x, s) = 2 F 2 c 2sc c 2π (k 2 + sc2 ) Finally, the inverse Laplace transform yields Green’s function with a source at the origin s 1 |x| 1 −1 1 , (1.11.73) exp − |x| = H t− G(x, t) = L 2c s c 2c c where H(z) is the Heaviside unit step function.
1.11 Green’s Functions and Boundary-Value Problems
With a point source at (ξ, τ ), Green’s function has the form |x − ξ| 1 . G(x, t; ξ, τ ) = H t − τ − 2c c
79
(1.11.74)
This function is also called the Riemann function for the wave equation. The result (1.11.74) shows that G = 0 unless the point (x, t) lies within the characteristic cone defined by the inequality c(t − τ ) > |x − ξ|. The solution of equation (1.11.67) is t ∞ dξ G(x, t; ξ, τ )q(ξ, τ ) dτ, u(x, t) = −∞
1 = 2c
0 ∞
t
dξ −∞
0
|x − ξ| q(ξ, τ ) dτ, H t−τ − c
(1.11.75)
which, since H = 1 for x − c(t − τ ) < ξ < x + c(t − τ ) and zero outside, t x+c(t−τ ) 1 1 = dτ q(ξ, τ ) dξ = q(ξ, τ ) dτ dξ, (1.11.76) 2c 0 2c x−c(t−τ ) D where D is the triangular domain made up of two points x ∓ ct on the x-axis and another point (x, t) off the x-axis in the (x, t)-plane. Thus, the solution of the general Cauchy problem described in Example 1.9.1 can be obtained by adding (1.11.75) to the d’Alembert solution (1.7.21), and hence, it reduces to (1.9.15). Example 1.11.7 (Green’s Function for an Axisymmetric Wave Equation). We consider the two-dimensional wave equation in polar coordinates
1 2 (1.11.77) − ∇ u − 2 utt = f (r, t), 0 < r < ∞, t > 0, c where the Laplacian ∇2 in cylindrical polar coordinates without θ-dependence is given by ∂ 1 ∂ r (1.11.78) ∇2 ≡ r ∂r ∂r and u = u(r, t) and f (r, t) are the given functions of r and t. We also prescribe appropriate initial conditions u(r, 0) = 0,
ut (r, 0) = 0.
Green’s function G(r, t) must satisfy the equation
1 1 2 δ(r)δ(t), − ∇ G − 2 Gtt = c 2πr with the same initial conditions (1.11.79).
(1.11.79)
(1.11.80)
80
1 Linear Partial Differential Equations
We apply the joint zero-order Hankel transform with respect to r and the Laplace transform with respect to t to (1.11.80) to obtain c2 , 2π(s2 + c2 κ2 )
˜ G(κ, s) =
(1.11.81)
where κ and s are the Hankel and Laplace transform variables, respectively. The inverse Laplace transform gives ˜ s) = G(κ,
c sin(cκt). 2πκ
(1.11.82)
Then, the inverse Hankel transform yields the solution for G(r, t) as c −1 1 G(r, t) = H sin(cκt) 2π 0 κ cH(t − rc ) cH(ct − r) = 1 = 2 1 . 2π(c2 t2 − r2 ) 2 2π(t2 − rc2 ) 2
(1.11.83)
This represents the Green function for the two-dimensional wave equation with a source at (0, 0). If this source is placed at the point r = ξ , the Green function satisfies the equation
1 (1.11.84) − ∇2 G − 2 Gtt = δ(x − ξ)δ(y − η)δ(t − τ ). c Introducing R = |r − ξ| and T = t − τ , we can rewrite (1.11.84) in the form
1 ∂ δ(R) ∂G 1 − R − 2 GT T = · δ(T ). (1.11.85) R ∂R ∂R c 2πR This is identical with (1.11.80), and hence, Green’s function is given by G(R, T ) =
cH(T − 2π(T 2 −
R c) R2 12 c2 )
=
|r−ξ| c } 1 2 12 c2 |r − ξ| }
cH{(t − τ ) − 2π{(t − τ )2 −
.
(1.11.86)
Example 1.11.8 (Green’s Function for the Three-Dimensional Inhomogeneous Wave Equation). The three-dimensional inhomogeneous wave equation is given by
1 (1.11.87) − ∇2 u − 2 utt = f (r, t), −∞ < x, y, z < ∞, t > 0, c where r = (x, y, z), and the Laplacian is given by ∇2 =
∂2 ∂2 ∂2 + 2 + 2. 2 ∂x ∂y ∂z
The initial and boundary conditions are
(1.11.88)
1.11 Green’s Functions and Boundary-Value Problems
u(r, 0) = 0, ut (r, t) → 0,
81
ut (r, t) = 0,
(1.11.89ab)
as r → ∞.
(1.11.90)
Green’s function G(x, t) for this problem satisfies the equation
1 − ∇2 G − 2 Gtt = δ(x)δ(y)δ(z)δ(t), −∞ < x, y, z < ∞, t > 0, c (1.11.91) with the same initial and boundary data (1.11.89ab), (1.11.90). Application of the joint Laplace and Fourier transform gives ˜ κ, s) = G(κ
c2 1 · 2 , 3/2 (s + c2 κ2 ) (2π)
κ = (k, , m).
The joint inverse transform yields the integral solution ∞ ∞ ∞ sin(cκt) c κ · x) dκ κ. exp(iκ G(x, t) = 3 (2π) −∞ −∞ −∞ κ
(1.11.92)
(1.11.93)
In terms of the spherical polar coordinates with the polar axis along the vector κ = κ2 dκ sin θ dθ dφ, integral (1.11.93) x, so that κ · x = κr cos θ, r = |x| and dκ assumes the form ∞ π 2π c dφ κ sin(cκt) dκ exp(iκr cos θ) sin θ dθ, G(x, t) = (2π)3 0 0 0 (1.11.94) ∞ iκr c = e − e−iκr sin(cκt) dκ 4π 2 ri 0 ∞ iκr c =− 2 e − e−iκr eicκt − e−icκt dκ 8π r 0 ∞ iκ(ct−r) c e = + e−iκ(ct−r) dκ 2 8π r 0
∞ iκ(ct+r) + e−iκ(ct+r) dκ e − 0
∞ ∞ c iκ(ct−r) iκ(ct+r) e dκ − e dκ = 8π 2 r −∞ −∞ 2πc δ(ct − r) − δ(ct + r) . (1.11.95) = 8π 2 r For t > 0, ct + r > 0, and hence, δ(ct + r) = 0. Thus, 1 r G(x, t) = , δ t− 4πr c in which the formula δ(ax) = a1 δ(x) is used.
(1.11.96)
82
1 Linear Partial Differential Equations
If the source is located at (ξ, η, ζ, τ ) = (ξξ , τ ), the desired Green function is given by
1 |x − ξ | |x − ξ | G(x, t; ξ , τ ) = δ t−τ − −δ t−τ + . 4π|x − ξ | c c (1.11.97) It should be noted that Green’s function (1.11.96) for the hyperbolic equation is a generalized function, whereas in the other examples of Green’s functions, it was always a piecewise analytic function. In general, Green’s function for an elliptic function is always analytic, whereas Green’s function for a hyperbolic equation is a generalized function.
1.12 Sturm–Liouville Systems and Some General Results We can generalize the method of separation of variables and the associated eigenvalue problems by considering the classical wave equation with variable coefficients ∂u ∂2u ∂ p(x) + qu = ρ 2 , 0 < x < , t > 0, (1.12.1) ∂x ∂x ∂t subject to the boundary condition for t > 0 a1 u + a2 u = 0,
b1 u + b2 u = 0,
x = 0;
x = ,
(1.12.2ab)
and the initial conditions u(x, 0) = f (x),
ut (x, 0) = g(x),
0 < x < ,
(1.12.3ab)
where p, q, and ρ are assumed to be continuous functions of x in 0 ≤ x ≤ and a1 , a2 , b1 , b2 are real positive constants such that a21 + a22 > 0
and
b21 + b22 > 0.
Using the method of separation of variables as in Example 1.6.1 with u(x, t) = X(x)T (t) = 0 and −λ as a separation constant, we obtain d dX p(x) + (q + λρ)X = 0, (1.12.4) dx dx d2 T + λT = 0, (1.12.5) dt2 with the boundary conditions a1 X + a2 X = 0,
x = 0;
b1 X + b2 X = 0,
where the prime denotes the derivative with respect to x.
x = , (1.12.6ab)
1.12 Sturm–Liouville Systems and Some General Results
83
The eigenvalue problem defined by (1.12.4) and (1.12.6ab) is called the Sturm– Liouville (SL) system. The values of λ for which the Sturm–Liouville problem has a nontrivial solution are called the eigenvalues, and the corresponding solutions are called the eigenfunctions. In terms of the operator
d d p(x) + q(x), (1.12.7) L= dx dx we can write (1.12.4) in the form, X(x) = u(x), Lu + λρu = 0.
(1.12.8)
The Sturm–Liouville equation (1.12.8) is called regular in a closed finite interval [a, b] if the functions p(x) and ρ(x) are positive in [a, b]. Thus, for a given λ, there exist two linearly independent solutions of a regular Sturm–Liouville equation (1.12.8) in [a, b]. The Sturm–Liouville equation (1.12.8) in [a, b] together with two separated end conditions a1 u(a) + a2 u (a) = 0,
b1 u(b) + b2 u (b) = 0,
(1.12.9ab)
where a1 , a2 , b1 , b2 are given real constants such that a21 + a22 > 0 and b21 + b22 > 0 is called a regular Sturm–Liouville (RSL) system. The set of all eigenvalues λ of a regular Sturm–Liouville problem is called the spectrum of the problem. Example 1.12.1. Consider the regular Sturm–Liouville problem u + λu = 0, 0 ≤ x ≤ π, u(0) = 0 = u(π).
(1.12.10) (1.12.11ab)
It is easy to check that, for λ ≤ 0, this problem has no nonzero solutions. In other words, there are no negative (λ < 0) or zero (λ = 0) eigenvalues of the problem. However, when λ > 0, then the solutions of the equation are √ √ u(x) = A cos λx + B sin λx. The boundary conditions (1.12.11ab) give √ A = 0 and B sin π λ = 0. Since λ = 0, and B = 0 yields a trivial solution, we must have B = 0, and hence, √ sin π λ = 0. Thus, the eigenvalues are λn = n2 , n = 1, 2, . . . , and the eigenfunctions are un (x) = sin nx. Note that λn → ∞ as n → ∞, unlike the case of self-adjoint, compact operators when the eigenvalues converge to zero (see Debnath and Mikusinski 1999, Theorem 4.9.9 and Theorem 5.10.4).
84
1 Linear Partial Differential Equations
Example 1.12.2. Consider the Cauchy–Euler equation x2 u + xu + λu = 0,
1≤x≤e
(1.12.12)
with the end conditions (1.12.13ab)
u(1) = 0 = u(e).
The Cauchy–Euler equation can be put into the Sturm–Liouville form du 1 d x + λu = 0. dx dx x The general solution of this equation is u(x) = C1 xi
√
λ
+ C2 x−i
√
λ
,
where C1 and C2 are arbitrary constants. In view of the fact that xia = exp(ia ln x) = cos(a ln x) + i sin(a ln x), the solution becomes u(x) = A cos
√
√ λ ln x + B sin λ ln x ,
where A and B are new arbitrary constants related to C1 and C2 . The end condition u(1) = 0 gives A = 0, and the end condition u(e) = 0 gives √ B = 0, sin λ = 0, which leads to the eigenvalues λn = (nπ)2 ,
n = 1, 2, 3, . . . ,
and the corresponding eigenfunctions un (x) = sin(nπ ln x),
n = 1, 2, 3, . . . .
A Sturm–Liouville equation (1.12.8) is called singular when it is given on a semiinfinite or infinite interval, or when the coefficient p(x) or ρ(x) vanishes, or when one of the coefficients becomes infinite at one end or both ends of a finite interval. A singular Sturm–Liouville equation together with appropriate linear homogeneous end conditions is called a singular Sturm–Liouville system. The conditions imposed in this case are not like the separated boundary conditions in the regular Sturm– Liouville problem.
1.12 Sturm–Liouville Systems and Some General Results
85
Example 1.12.3. We consider the singular Sturm–Liouville problem involving Legendre’s equation d 1 − x2 u + λu = 0, dx
−1 < x < 1,
(1.12.14)
with the boundary conditions that u and u are finite as x → ±1. In this case, p(x) = 1 − x2 and ρ(x) = 1, and p(x) vanishes at x = ±1. The Legendre functions of the first kind Pn (x), n = 0, 1, 2, . . . , are the eigenfunctions which are finite as x → ±1. The corresponding eigenvalues are λn = n(n + 1) for n = 0, 1, 2, . . . . We note that the singular Sturm–Liouville problem has infinitely many eigenvalues, and the eigenfunctions Pn (x) are orthogonal to each other with respect to the weight function ρ(x) = 1. Example 1.12.4. Another example of a singular Sturm–Liouville problem is the Bessel equation for fixed ν du ν2 d x + λx − u = 0, 0 < x < a, (1.12.15) dx dx x with the end conditions that u(a) = 0 and u, u are finite as x → 0+. 2 In this case, p(x) = x, q(x) = − νx , and ρ(x) = x. Here p(0) = 0, q(x) is infinite as x → 0+, and ρ(0) = 0. Therefore, the problem is singular. If λ = k2 , the eigenfunctions are the Bessel functions Jν (kn x) of the first kind of order ν where n = 1, 2, 3, . . . , and (kn a) is the nth zero of Jν . The eigenvalues are λn = kn2 . The Bessel function Jν and its derivative are both finite as x → 0+. Thus, the problem has infinitely many eigenvalues and the eigenfunctions are orthogonal to each other with respect to the weight function ρ(x) = x. In the preceding examples, we see that the eigenfunctions are orthogonal with respect to the weight function ρ(x). In general, the eigenfunctions of a singular SL system are orthogonal with respect to the weight function ρ(x), which will be proved later on. Another type of problem that often arises in practice is the periodic Sturm– Liouville system: d du p(x) + (q + λρ)u = 0, a ≤ x ≤ b, (1.12.16) dx dx in which p(a) = p(b), together with the periodic end conditions u(a) = u(b),
u (a) = u (b).
(1.12.17ab)
Example 1.12.5. Find the eigenvalues and eigenfunctions of the periodic Sturm– Liouville system: u + λu = 0, u(−π) = u(π),
−π ≤ x ≤ π, u (π) = u (−π).
(1.12.18) (1.12.19ab)
86
1 Linear Partial Differential Equations
Note that here p(x) = 1 and hence p(π) = p(−π). For λ > 0, the general solution of the equation is √ √ u(x) = A cos λx + B sin λx, where A and B are arbitrary constants. Using the boundary conditions (1.12.19ab), we obtain √ 2B sin λπ = 0, √ √ 2A λ sin λπ = 0. Thus, for nontrivial solutions, we must have √ sin λπ = 0, A = 0,
B = 0.
Consequently, λn = n2 ,
n = 1, 2, 3, . . . .
So, for every eigenvalue λn = n2 , there are two linearly independent solutions cos nx and sin nx. It can easily be checked that there are no negative eigenvalues of the system. However, λ = 0 is an eigenvalue and the associated eigenfunction is the constant function u(x) = 1. Thus, the eigenvalues are 0, {n2 }, and the corresponding eigenfunctions are 1, {cos nx}, {sin nx}, where n is a positive integer. For the regular Sturm–Liouville problem, we denote the domain of L by D(L), that is, D(L) is the space of all complex-valued functions u defined on [a, b] for which u ∈ L2 ([a, b]) and which satisfy boundary conditions (1.12.9ab). Theorem 1.12.1 (Lagrange’s Identity). For any u, v ∈ D(L), we have uLv − vLu =
d p uv − vu . dx
(1.12.20)
Proof. We have dv d du d p + quv − v p − quv uLv − vLu = u dx dx dx dx d = p uv − vu . dx Theorem 1.12.2 (Abel’s Formula). If u and v are two solutions of the equation (1.12.8) in [a, b], then p(x)W (u, v; x) = const., where W is the Wronskian defined by W (u, v; x) = uv − u v .
(1.12.21)
1.12 Sturm–Liouville Systems and Some General Results
87
Proof. Since u, v are solutions of (1.12.8), we have d pu + (q + λρ)u = 0, dx d pv + (q + λρ)v = 0. dx Multiplying the first equation by v and the second equation by u, and then subtracting gives d d pv − v pu = 0. dx dx Integrating this equation from a to x yields p(x) u(x)v (x) − u (x)v(x) − p(a) u(a)v (a) − u (a)v(a) = 0. u
This is Abel’s formula. Theorem 1.12.3. The Sturm–Liouville operator L is self-adjoint. In other words, for any u, v ∈ D(L), we have Lu, v = u, Lv, where ·, · denotes the inner product in L2 ([a, b]) defined by b f (x)g(x) dx. f, g =
(1.12.22)
(1.12.23)
a
Proof. Since all constants involved in the boundary conditions of a Sturm–Liouville system are real, if v ∈ D(L), then v ∈ D(L). Also since p, q and ρ are real valued, Lv = Lv. Consequently, we have b (vLu − uLv) dx Lu, v − u, Lv = a
b = p vu − uv a ,
by Lagrange’s identity (1.12.20). (1.12.24)
We shall show that the right-hand side of the above equality vanishes for both the regular and singular SL systems. If p(a) = 0, the result follows immediately. If p(a) > 0, then u and v satisfy the boundary conditions of the form (1.12.9ab) at x = a. That is,
a1 u(a) u (a) = 0. v(a) v (a) a2 Since a1 and a2 are not both zero, we have v(a)u (a) − u(a)v (a) = 0. A similar argument can be used to the other end point x = b, so that the right-hand side of (1.12.24) vanishes. This proves the theorem.
88
1 Linear Partial Differential Equations
Theorem 1.12.4. All eigenvalues of a Sturm–Liouville system are real. Proof. Let λ be an eigenvalue of an SL system and let u(x) be the corresponding eigenfunction. This means that u = 0 and Lu = −λρu. Then
b
0 = Lu, u − u, Lu = (λ − λ)
2 ρ(x)u(x) dx.
a
Since ρ(x) > 0 in [a, b] and u = 0, the integral is a positive number. Thus λ = λ. This completes the proof. Remark. This theorem states that all eigenvalues of a regular SL system are real, but it does not guarantee that an eigenvalue exists. It has been shown by example that an SL system has an infinite sequence of eigenvalues. All preceding examples of the SL system suggest that λ1 < λ2 < λ3 < · · · with lim λn = ∞.
n→∞
All of these results can be stated in the form of a theorem as follows. Theorem 1.12.5. The eigenvalues λn of an SL system can be arranged in the form λ1 < λ2 < λ3 < · · · , and lim λn = ∞,
n→∞
(1.12.25)
so that n refers to the number of zeros of the eigenfunctions un (x) in [a, b]. The proof of this theorem is beyond the scope of this book, and we refer to Debnath and Mikusinski (2005). Theorem 1.12.6. The eigenfunctions corresponding to distinct eigenvalues of a Sturm–Liouville system are orthogonal with respect to the inner product with the weight function ρ(x). Proof. Suppose u1 (x) and u2 (x) are eigenfunctions corresponding to eigenvalues λ1 and λ2 with λ1 = λ2 . Thus, Lu1 = −λ1 ρu1
and
Lu2 = −λ2 ρu2 .
Hence u1 Lu2 − u2 Lu1 = (λ1 − λ2 )ρu1 u2 . By Theorem 1.12.1, we have
(1.12.26)
1.12 Sturm–Liouville Systems and Some General Results
u1 Lu2 − u2 Lu1 =
d p u1 u2 − u2 u1 . dx
89
(1.12.27)
Combining (1.12.26) and (1.12.27) and integrating from a to b gives b ρ(x)u1 (x)u2 (x) dx (λ1 − λ2 ) a
b = p(x) u1 (x)u2 (x) − u2 (x)u1 (x) a = 0, by boundary conditions (1.12.9ab). Since λ1 = λ2 , this equality shows that b ρ(x)u1 (x)u2 (x) dx = 0. a
This proves the theorem. We consider some general results about the eigenfunction expansions and their completeness property of an SL system. Suppose {un (x)}∞ n=1 is a set of orthogonal eigenfunctions of an SL system in [a, b]. The inner product of these functions with respect to the weight function ρ(x) is defined by b un , um = ρ(x)un (x)um (x) dx, (1.12.28) a
so that the square of the norm is un 2 = un , un =
b
ρ(x)u2n (x) dx.
(1.12.29)
a
The set of orthogonal eigenfunctions of an SL system is said to be complete if any arbitrary function f ∈ L2 ([a, b]) can be expanded uniquely as f (x) =
∞
an un (x),
(1.12.30)
n=1
where the series converges to f (x) in L2 ([a, b]) and the coefficients an are given by an =
1 f (x), un (x) = f, un , un (x), un (x) un 2
(1.12.31)
where n = 1, 2, 3, . . . . The expansion (1.12.30) is called a generalized Fourier series of f (x) and the associated scalars an are called the generalized Fourier coefficients of f (x). The set of eigenfunctions {un }∞ n=1 is called orthonormal if un = 1. Obviously, the set of orthonormal eigenfunctions is said to be complete, if, for every f ∈ L2 ([a, b]), the following expansion holds:
90
1 Linear Partial Differential Equations
f (x) =
∞
a n un =
n=1
∞
f, un un .
(1.12.32)
n=1
A series of orthonormal eigenfunctions f (x) in L2 ([a, b]) if
∞ n=1
an un (x) is said to be convergent to
lim f (x) − sn (x) = 0,
n→∞
(1.12.33)
n where sn (x) = r=1 ar ur (x) is the nth partial sum of series (1.12.32). Equivalently, (1.12.33) reads as 2 b n ar ur (x) ρ(x) dx = 0. lim f (x) − n→∞ a
(1.12.34)
r=1
This type of convergence is called the strong convergence and is entirely different from pointwise or uniform convergence in analysis. In general, the strong convergence in L2 ([a, b]) implies neither pointwise convergence nor uniform convergence. However, the uniform convergence implies both strong convergence and pointwise convergence. We now determine the coefficients ar such that the nth partial sum sn (x) of the series (1.12.32) represents the best approximation to f (x) in the sense of least squares, that is, we seek to minimize the integral in (1.12.34) b f (x) −
I(ar ) = a
n
2 ar ur (x)
b
ρ(x)f (x) dx − 2 2
= a
+
ρ(x) dx
r=1
n
n
r=1
b
ar
ρ(x)f (x)ur (x) dx a
b
a2r
ρ(x)u2r (x) dx.
(1.12.35)
a
r=1
This is an extremal problem. A necessary condition for I(ar ) to be minimum is that the first partial derivatives of I with respect to the coefficients ar vanish. Thus, we obtain b b n ∂I 2 = ρur f (x) dx + 2ar ρur dx = 0. (1.12.36) −2 ∂ar a a r=1 Consequently,
b
f (x)ur (x)ρ(x) dx = f, ur .
ar = a
If we complete the square, the right-hand side of (1.12.35) becomes
(1.12.37)
1.12 Sturm–Liouville Systems and Some General Results
b 2
I(ar ) =
ρf dx + a
n
2
b
ar −
ρf ur dx a
r=1
−
n r=1
b
91
2 ρf ur dx .
a
(1.12.38) The right-hand side shows that I is minimum if and only if ar is given by (1.12.37). This choice of ar gives the best approximation to f (x) in the sense of least squares. Substituting the values of ar into (1.12.35) gives 2 b b n n f (x) − ar ur (x) ρ(x) dx = ρ(x)f 2 (x) dx − a2r . (1.12.39) a
a
r=1
r=1
If the series of orthonormal eigenfunctions converges to f (x), then (1.12.34) is satisfied. Invoking the limit as n → ∞ in (1.12.39) and using (1.12.34) gives the Parseval relation b ∞ a2r = ρ(x)f 2 (x) dx = f 2 , (1.12.40) a
r=1
or equivalently, ∞ f, ur 2 = f 2 .
(1.12.41)
r=1
Since the left-hand side of (1.12.39) is nonnegative, it follows from (1.12.39) that n
a2r ≤ f 2 .
(1.12.42)
r=1
Since the right-hand side of (1.12.42) is finite, the left-hand side of (1.12.42) is bounded above for any n. Proceeding to the limit as n → ∞ gives the inequality ∞
a2n ≤ f 2 ,
(1.12.43)
n=1
or equivalently, ∞ f, un 2 ≤ f 2 .
(1.12.44)
n=1
This is called Bessel’s inequality. In Section 1.6, the method of separation of variables or the Fourier method or the method of eigenfunction expansions has been discussed with many examples. This is the basic method for solving partial differential equations in bounded special domains. We now illustrate the generalized Fourier method by solving more general Sturm–Liouville problems associated with a general wave equation and a general diffusion equation.
92
1 Linear Partial Differential Equations
Example 1.12.6 (Solution of the Sturm–Liouville Problem Associated with the Wave Equation). We develop the generalized Fourier method by solving a more general Sturm–Liouville equation associated with the wave equation
1 ∂u ∂ ∂2u = p + qu + F (x, t), a ≤ x ≤ b, t > 0, (1.12.45) ∂t2 ρ(x) ∂x ∂x with the boundary conditions (1.12.2ab) and the initial conditions (1.12.3ab), where F (x, t) is the forcing (source) term. In terms of the SL operator, equation (1.12.45) can be written as a ≤ x ≤ b, t > 0,
utt = Lu + F (x, t), where Lu =
∂ 1 (pux ) + qu . ρ(x) ∂x
(1.12.46)
(1.12.47)
Following the method of separation of variables, we assume the solution of the wave equation (1.12.45) with F = 0 in the form u(x, t) = φ(x)ψ(t) = 0 so that equation (1.12.45) reduces to d2 ψ = λψ, dt2 Lφ = λφ,
t > 0,
(1.12.48)
a ≤ x ≤ b,
(1.12.49)
where λ is a separation constant. The associated boundary conditions for φ(x) are a1 φ(a) + a2 φ (a) = 0,
b1 φ(b) + b2 φ (b) = 0.
(1.12.50ab)
Equation (1.12.49) with (1.12.50ab) is called the associated SL problem. In general, this problem can be solved by finding the eigenvalues λn and the orthonormal eigenfunctions φn (x), n = 1, 2, 3, . . . . Using the principle of superposition, we can write the solution of the linear equation (1.12.46) in the form u(x, t) =
∞
φn (x)ψn (t),
(1.12.51)
n=1
where ψn (t) are to be determined. We further assume that the forcing term can also be expanded in terms of the eigenfunctions as F (x, t) =
∞
φn (x)fn (t),
n=1
where the generalized Fourier coefficients fn (t) of F (x, t) are given by
(1.12.52)
1.12 Sturm–Liouville Systems and Some General Results
93
b
fn (t) = F, φn =
F (x, t)φn (x) dx.
(1.12.53)
a
Substituting (1.12.51) and (1.12.52) into (1.12.46) gives ∞ ∞ ∞ ¨ φn (x)ψn (t) + φn (x)fn (t) ψn (t)φn (x) = L n=1
n=1 ∞
= =
n=1
ψn (t)Lφn (x) +
n=1 ∞
∞
φn (x)fn (t)
n=1
λn ψn (t) + fn (t) φn (x).
n=1
This leads to the ordinary differential equation ψ¨n (t) + α2 ψn (t) = fn (t), n
(1.12.54)
where λn = −αn2 . Application of the Laplace transform method leads to the solution of (1.12.54) as 1 ˙ ψ(0) sin(αn t) ψn (t) = ψn (0) cos(αn t) + α n t 1 sin αn (t − τ )fn (τ ) dτ, (1.12.55) + αn 0 where ψn (0) and ψ˙ n (0) can be determined from the initial data (1.12.3ab) so that ∞ φn (x)ψn (0), (1.12.56) f (x) = u(x, 0) = n=1 ∞
g(x) = ut (x, 0) =
ψ˙ n (0)φn (x),
(1.12.57)
n=1
which give the generalized Fourier coefficients ψn (0) and ψ˙ n (0) as follows: b f (ξ)φn (ξ) dξ, (1.12.58) ψn (0) = f, φn = ab g(ξ)φn (ξ) dξ. (1.12.59) ψ˙ n (0) = g, φn = a
Therefore, the final solution is given by ∞ u(x, t) = ψn (t)φn (x) =
n=1 ∞
1 g, φn sin αn t αn n=1
t 1 sin αn (t − τ )fn (τ ) dτ φn (x). + αn 0 f, φn cos αn t +
(1.12.60)
94
1 Linear Partial Differential Equations
This represents an infinite series solution of the wave equation (1.12.45) with the boundary and initial data (1.12.2ab) and (1.12.3ab) under appropriate conditions on the initial data f (x), g(x) and the forcing term F (x, t). Replacing the inner products by the integrals (1.12.58) and (1.12.59), and fn (τ ) by (1.12.53) and interchanging the summation and integration, we obtain the solution in the form b ∞ u(x, t) = φn (x)φn (ξ) cos αn t f (ξ) dξ a n=1 b ∞ 1 + φn (x)φn (ξ) sin αn t g(ξ) dξ a n=1 αn b t ∞ 1 φn (x)φn (ξ) sin αn (t − τ ) F (ξ, τ ) dξ dτ. (1.12.61) + a 0 n=1 αn Example 1.12.7 (Solution of the Sturm–Liouville Problem Associated with the Diffusion Equation). We consider the diffusion equation with a forcing (source) term F (x, t) in the form
∂u ∂ ∂u p(x) + q(x)u + F (x, t), a ≤ x ≤ b, t > 0, (1.12.62) = ∂t ∂x ∂x with boundary conditions a1 u(a, t) + a2 ux (a, t) = 0,
b1 u(b, t) + b2 ux (b, t) = 0,
t > 0,
(1.12.63ab)
and the initial condition u(x, 0) = f (x),
a < x < b.
(1.12.64)
In terms of the SL operator L, equation (1.12.62) takes the form ut = Lu + F.
(1.12.65)
We use the method of separation of variables to seek a solution of the equation (1.12.62) with F = 0 in the form u(x, t) = φ(x)ψ(t) = 0 so that the equation (1.12.62) becomes dψ = λψ, dt Lφ = λφ,
t > 0,
(1.12.66)
a ≤ x ≤ b,
(1.12.67)
where λ is the separation constant. The associated boundary conditions are a1 φ(a) + a2 φ (a) = 0,
b1 φ(b) + b2 φ (b) = 0.
(1.12.68ab)
Equation (1.12.67) with (1.12.68ab) is called the associated SL problem, which can easily be solved by finding the eigenvalues λn and the orthonormal eigenfunctions
1.12 Sturm–Liouville Systems and Some General Results
95
φn (x), n = 1, 2, 3, . . . . According to the linear superposition principle, we write the solution of (1.12.67) in the form u(x, t) =
∞
(1.12.69)
φn (x)ψn (t),
n=1
where ψn (t) are to be determined. We further assume that the forcing function can also be expanded in terms of the eigenfunctions as F (x, t) =
∞
(1.12.70)
fn (t)φn (x),
n=1
where the Fourier coefficients fn (t) are given by
b
fn (t) = F, φ =
F (ξ, t)φn (ξ) dξ.
(1.12.71)
a
Putting (1.12.69) and (1.12.70) in equation (1.12.65) yields ∞ ∞ ∞ ˙ φn (x)ψn (t) = L φn (x)ψn (t) + fn (t)φn (x) n=1
n=1
n=1
∞ = ψn (t)Lφn (x) + fn (t)φn (x)
=
n=1 ∞
λn ψn (t) + fn (t) φn (x).
n=1
This gives an ordinary differential equation for ψn (t) as ψ˙ n (t) = λn ψn (t) + fn (t). Applying the Laplace transform to this equation gives the solution t ψn (t) = ψn (0) exp(λn t) + exp λn (t − τ )fn (τ ) dτ ,
(1.12.72)
(1.12.73)
0
where n = 1, 2, 3, . . . and ψn (0) can be determined from the initial condition f (x) = u(x, 0) =
∞
φn (x)ψn (0),
(1.12.74)
n=1
where the Fourier coefficients ψn (0) of the function f (x) are given by b f (ξ)φn (ξ) dξ. ψn (0) = f, φn = a
Substituting (1.12.73) in the solution (1.12.69) gives
(1.12.75)
96
1 Linear Partial Differential Equations
u(x, t) =
∞
t
f, φn exp(λn t) +
exp λn (t − τ )fn (τ ) dτ φn (x).
0
n=1
(1.12.76) We next replace the inner product in (1.12.76) by (1.12.75), fn (τ ) by (1.12.71) and interchange the summation and integration to obtain the final form of the solution in the form b ∞ u(x, t) = φn (ξ)φn (x) exp(λn t) f (ξ) dξ a
+
n=1 ∞ t b 0
φn (ξ)φn (x) exp λn (t − τ ) F (ξ, τ ) dξ dτ.
a n=1
(1.12.77) Introducing a new function G defined by G(x, ξ, t) =
∞
φn (ξ)φn (x)eλn t ,
(1.12.78)
n=1
we can write the solution in terms of G in the form b G(x, ξ, t)f (ξ) dξ u(x, t) = a
t
b
G(x, ξ, t − τ )F (ξ, τ ) dξ dτ.
+ 0
(1.12.79)
a
It is noted that the first term of this solution represents the contribution from the initial condition, and the second term is due to the nonhomogeneous term of the equation (1.12.62). A typical boundary-value problem for an ordinary differential equation can be written in the operator form as Lu = f,
a ≤ x ≤ b.
(1.12.80)
Usually, we seek a solution of this equation with the given boundary conditions. One formal approach to the problem is to find the inverse operator L−1 . Then the solution of (1.12.80) can be found as u = L−1 (f ). It turns out that it is possible in many important cases, and the inverse operator is an integral operator of the form u(x) = L−1 f (x) =
b
G(x, t)f (t) dt.
(1.12.81)
a
The function G is called Green’s function of the operator L. The existence of Green’s function and its construction is not a simple problem in the case of the regular Sturm– Liouville system.
1.12 Sturm–Liouville Systems and Some General Results
97
Theorem 1.12.7 (Green’s Function for an SL System). Suppose λ = 0 is not an eigenvalue of the following regular SL system: du d p(x) + q(x)u = f (x), a ≤ x ≤ b, (1.12.82) Lu ≡ dx dx with the homogeneous boundary conditions a1 u(a) + a2 u (a) = 0,
b1 u(b) + b2 u (b) = 0,
(1.12.83ab)
where p and q are continuous real-valued functions on [a, b], p is positive in [a, b], p (x) exists and is continuous in [a, b], and a1 , a2 , b1 , b2 , are given real constants such that a21 + a21 > 0 and b21 + b21 > 0. Thus, for any f ∈ C 2 ([a, b]), the SL system has a unique solution b u(x) = G(x, t)f (t) dt, (1.12.84) a
where G is the Green function given by ⎧ ⎨ u2 (x)u1 (t) p(t)W (t) G(x, t) = u (x)u (t) 2 ⎩ 1 p(t)W (t)
if a ≤ t < x, if x < t ≤ b,
(1.12.85)
where u1 and u2 are nonzero solutions of the homogeneous system (f = 0) and W is the Wronskian given by u1 (t) u2 (t) . W (t) = u1 (t) u2 (t) Proof. According to the theory of ordinary differential equations, the general solution of (1.12.82) is of the form u(x) = c1 u1 (x) + c2 u2 (x) + up (x),
(1.12.86)
where c1 and c2 are arbitrary constants, u1 and u2 are two linearly independent solutions of the homogeneous equation Lu = 0, and up (x) is any particular solution of (1.12.82). Using the method of variation of parameters, we obtain the particular solution up (x) = u1 (x)v1 (x) + u2 (x)v2 (x), where v1 (x) and v2 (x) are given by f (x)u2 (x) v1 (x) = − dx, p(x)W (x)
v2 (x) =
f (x)u1 (x) dx. p(x)W (x)
(1.12.87)
(1.12.88)
According to Abel’s formula (see Theorem 1.12.2), p(x)W (x) is a constant. Since W (x) = 0 in [a, b] and p(x) is assumed to be positive, the constant is nonzero. Denoting the constant by c so that
98
1 Linear Partial Differential Equations
1 , p(x)W (x)
c= then
v1 (x) = −
and v2 (x) =
cf (x)u2 (x) dx
cf (x)u1 (x) dx.
Thus, the final form of up (x) is x f (t)u2 (t) dt + cu2 (x) up (x) = −cu1 (x)
b
x
f (t)u1 (t) dt
a b
cu2 (x)u1 (t)f (t) dt +
=
x
cu1 (x)u2 (t)f (t) dt.
a
(1.12.89)
x
Consequently, if we denote Green’s function as cu2 (x)u1 (t) if a ≤ t < x, G(x, t) = cu1 (x)u2 (t) if x < t ≤ b, we can write
(1.12.90)
b
up (x) =
G(x, t)f (t) dt,
(1.12.91)
a
provided the integral exists. This follows immediately from the continuity of G. The continuity of G is left as an exercise. We denote the integral operator T given by (1.12.81), that is, b G(x, t)f (t) dt. (1.12.92) (T f )(x) = a
Theorem 1.12.8. The operator T defined by (1.12.92) is self-adjoint from L2 ([a, b]) into C([a, b]) if G(x, t) = G(t, x). Proof. The function G(x, t) defined on [a, b] × [a, b] is continuous if b b G(x, t)2 dx dt < ∞. a
We have
a
b
b
T f, g =
G(x, t)f (t)g(x) dx dt a
b
a b
=
G(x, t)f (t)g(x) dx dt
a b
a b
f (t) dt $ a a∗ % = f, T g ,
b
G(x, t)g(x) dx
=
a
1.12 Sturm–Liouville Systems and Some General Results
which shows that
T ∗ f (x) =
99
b
G(t, x)f (t) dt. a
Obviously, T is self-adjoint if its kernel satisfies the equality G(x, t) = G(t, x). Theorem 1.12.9. Under the assumptions of Theorem 1.12.7, λ is an eigenvalue of L if and only if (1/λ) is an eigenvalue of T . Furthermore, if f is an eigenfunction of L corresponding to the eigenvalue λ, then f is an eigenfunction of T corresponding to the eigenvalue (1/λ). Proof. Suppose Lf = λf for some nonzero f in the domain of L. In view of the definition of T , and Theorem 1.12.7, we have f = L−1 (λf ) = T (λf ). Or equivalently, since λ = 0, Tf =
1 f. λ
This means (1/λ) is an eigenvalue of T with the corresponding eigenfunction f . Conversely, if (f = 0) is an eigenfunction of T corresponding to the eigenvalue λ = 0, then T f = λf. Since T = L−1 , one has f = L(T f ) = L(λf ) = λL(f ). Thus, (1/λ) is an eigenvalue of L and the corresponding eigenfunction is f . Theorem 1.12.10 (Bilinear Expansion of Green’s Function). If G(x, t) is Green’s function for the regular SL system (1.12.82), (1.12.83ab) and the associated eigenvalue problem Lφ = λφ,
a ≤ x ≤ b,
(1.12.93)
with (1.12.83ab) has infinitely many nonzero eigenvalues λn with the corresponding orthonormal eigenfunctions φn , then G(x, t) can be expanded in terms of λn and φn as ∞ 1 G(x, t) = φn (x)φn (t). (1.12.94) λn n=1 Proof. We assume that the solution u(x) of (1.12.82), (1.12.83ab) is given in terms of the eigenfunctions as
100
1 Linear Partial Differential Equations
u(x) =
∞
(1.12.95)
an φn (x),
n=1
where the coefficients an are to be determined. We next express the given forcing function f in terms of the eigenfunctions as f (x) =
∞
(1.12.96)
fn φn (x),
n=0
where the coefficients fn are fn = f, φn =
b
f (t)φn (t) dt.
(1.12.97)
a
Substituting (1.12.95), (1.12.96) into (1.12.82) yields ∞ ∞ an φn (x) = fn φn (x). L n=1
But the left-hand side of (1.12.98) is ∞ ∞ ∞ L an φn (x) = an L φn (x) = an λn φn (x). n=1
(1.12.98)
n=1
n=1
(1.12.99)
n=1
Equating the right-hand side of (1.12.98) and (1.12.99) yields 1 1 1 fn = f, φn = an = λn λn λn
b
f (t)φn (t) dt.
(1.12.100)
a
Consequently, (1.12.95) leads to the result b
∞ 1 u(x) = f (t)φn (t) dt φn (x), λ a n=1 n which, by interchanging the summation and integration, b ∞ 1 φn (t)φn (x) f (t) dt. = a n=1 λn
(1.12.101)
In view of (1.12.84), Green’s function G(x, t) is then given by G(x, t) = This is the desired result (1.12.94).
∞ 1 φn (t)φn (x). λ n=1 n
(1.12.102)
1.13 Energy Integrals and Higher Dimensional Equations
101
1.13 Energy Integrals and Higher Dimensional Equations In this section, we discuss the energy integrals, the law of conservation of energy, uniqueness theorems, higher dimensional wave equations, and diffusion equations. We first derive the energy integral associated with the (1 + 1)-dimensional wave equation problem utt = c2 uxx ,
a ≤ x ≤ b, t > 0,
u(a, t) = 0 = u(b, t) u(x, 0) = f (x),
t > 0,
ut (x, 0) = g(x),
a ≤ x ≤ b.
(1.13.1) (1.13.2ab) (1.13.3ab)
Multiplying equation (1.13.1) by ut , we can rewrite the result in the form d 1 2 1 2 2 ∂ ut + c ux − c2 (ut ux ) = 0. (1.13.4) dt 2 2 ∂x Integrating this equation with respect to x from a to b gives d dt
b a
1 2 ut + c2 u2x dx = c2 [ut ux ]ba 2 = c2 ut (b, t)ux (b, t) − ut (a, t)ux (a, t) ,
(1.13.5)
which is zero because ut (a, t) = 0 = ut (b, t), which follows from (1.13.2ab). We introduce b 1 2 ut + c2 u2x dx, (1.13.6) E(t) = a 2 which is called the energy integral or the total energy of the system. It follows from (1.13.5) that dE = 0. dt
(1.13.7)
E(t) = const. = E0 .
(1.13.8)
This implies that
This means that the energy of the wave equation system is conserved. The energy equation (1.13.8) can be used to prove the uniqueness theorem which states that the wave equation system (1.13.1)–(1.13.3ab) has a unique solution. Suppose that there are two solutions u(x, t) and v(x, t) of the system and set w(x, t) = u − v. Obviously, w(x, t) satisfies the following equation: wtt = c2 wxx ,
a ≤ x ≤ b, t > 0,
102
1 Linear Partial Differential Equations
w(a, t) = 0 = w(b, t), t > 0, w(x, 0) = 0 = wt (x, 0), a < x < b. In view of equation (1.13.5) with u = w and the law of conservation of energy (1.13.8), we obtain d b 2 dE = wt + c2 wx2 dx = 0. dt dt a Since the integrand is positive, wt = 0 = wx for all x and t, hence w ≡ 0, which means that u = v. This proves the uniqueness. We next consider the (n + 1)-dimensional wave equation in the form 2 ∂2u ∂2u ∂2u 2 ∂ u = c + + · · · + (1.13.9) = c2 ∇2n u, ∂t2 ∂x21 ∂x22 ∂x2n where x = (x1 , x2 , . . . , xn ) ∈ Rn , t > 0, and ∇2n is the n-dimensional Laplacian. Suppose e = (e1 , e2 , . . . , en ) is a unit vector in Rn so that e21 + e22 + · · · + e2n = 1.
(1.13.10)
For a fixed t and a constant a, the equation e1 x1 + e2 x2 + · · · + en xn − ct = a
(1.13.11)
represents a plane in the x-space Rn . The unit vector e is normal to the above plane. As t increases, the plane moves in the direction of e with constant speed c. It is easy to verify that u(x, t) = F (x · e − ct)
(1.13.12)
is a solution of the wave equation (1.13.9). The value of u on the plane (1.13.11) is equal to the constant F (a). Usually, solutions (1.13.12) of the wave equation (1.13.9) are called plane waves. This idea of plane waves is consistent with that of the wave equation for n = 1, 2, 3. For n = 1 with x1 = x, equation (1.13.9) gives the (1 + 1)-dimensional wave equation utt = c2 uxx . Obviously, condition (1.13.10) becomes e21 = 1 which gives only two possible values of e1 = ±1. Thus, the corresponding plane wave solutions are u(x, t) = F (x − ct) and u(x, t) = G(x + ct), where F and G are arbitrary C 2 functions of a single variable. The former represents a wave traveling in the positive x-direction with constant speed c and the latter also represents a wave traveling in the negative x-direction with constant speed c. For n = 2 with x1 = x, and x2 = y, (1.13.9) become the (2 + 1)-dimensional wave equation
1.13 Energy Integrals and Higher Dimensional Equations
103
utt = c2 (uxx + uyy ). In this case, there are infinitely many unit vectors (e1 , e2 ) in R2 . For example, e = ( √12 , √12 ) is a unit vector and the corresponding plane wave solutions are 1 1 u(x, y, t) = F √ x + √ y − ct , 2 2
(1.13.13)
where F is an arbitrary C 2 function of a single variable. For n = 3 with x1 = x, x2 = y, x3 = z, equation (1.13.9) represents the (3 + 1)-dimensional wave equation utt = c2 (uxx + uyy + uzz ). Again there are infinitely many unit vectors e = (e1 , e2 , e3 ). For example, one such vector is e = (1, 0, 0) and the corresponding plane wave solution is u(x, y, z, t) = F (x − ct). In cylindrical polar coordinates (x1 = r cos θ, x2 = r sin θ, z = z), the (3 + 1)dimensional wave equation is
∂2u ∂u 1 ∂ 2u ∂2u 2 1 ∂ = c + . (1.13.14) r + ∂t2 r ∂r ∂r r 2 ∂θ2 ∂z 2 If u is independent of θ and z, this wave equation reduces to
∂2u ∂u 2 1 ∂ r . =c ∂t2 r ∂r ∂r
(1.13.15)
This is known as the cylindrical wave equation. In spherical polar coordinates (x = r sin φ cos θ, y = r sin φ sin θ, z = r cos φ) in R3 , the (3 + 1)-dimensional wave equation (1.13.9) takes the form
1 ∂ ∂u 1 ∂2u ∂2u c2 ∂ 2 ∂u , (1.13.16) r + sin φ + = 2 ∂t2 r ∂r ∂r sin φ ∂φ ∂φ sin2 φ ∂θ2 where 0 ≤ r ≤ ∞, 0 ≤ θ ≤ 2π, and 0 ≤ φ ≤ π. If u does not depend on θ and φ, then (1.13.16) reduces to the form c2 ∂ ∂2u 2 ∂u = 2 r . (1.13.17) ∂t2 r ∂r ∂r This is known as the equation of spherical waves. Substituting w = ru in (1.13.17) leads to the one-dimensional wave equation ∂2w ∂2w = c2 2 . 2 ∂t ∂r
(1.13.18)
104
1 Linear Partial Differential Equations
Every solution of this equation can be expressed as the sum of solutions of the form F (r − ct) and G(r + ct) where F and G are arbitrary C 2 functions of a single variable. Obviously, solutions of (1.13.17) can be written as the sum of solutions of the form u(r, t) =
1 F (r − ct) r
and u(r, t) =
1 G(r + ct). r
(1.13.19)
These are known as spherical waves. In general, the wave equation (1.13.9) can be expressed in spherical coordinates in the x-space Rn in the form
∂2u 1 ∂ 1 2 n−1 ∂u = c Δ u , (1.13.20) r + n ∂t2 r n−1 ∂r ∂r r2 where Δn is a second-order partial differential operator involving derivatives with respect to the angular coordinates. According to usual definitions, a spherical wave is a solution of (1.13.20) which depends only on r and t and does not depend on angular coordinates. Thus, the equation of spherical waves is
1 ∂ ∂ 2u 2 n−1 ∂u =c r . (1.13.21) ∂t2 r n−1 ∂r ∂r We can apply the method of separation of variables to the wave equation (1.13.9) by expressing a solution of the form u(x, t) = v(x)T (t). Substituting this separable solution into (1.13.9) gives
T¨(t) ∂2v ∂2v 1 ∂2v = 2 = −λ, (1.13.22) + + ··· + 2 2 2 v(x) ∂x1 ∂x2 ∂xn c T (t) where −λ is a separation constant. This leads to the pair of equations ∂2v ∂2v ∂ 2v + + · · · + + λv = 0, ∂x21 ∂x22 ∂x2n T¨ + λc2 T = 0.
(1.13.23) (1.13.24)
Equation (1.13.23) is a famous equation of the elliptic type and is often known as the reduced wave equation. We consider the first important case λ = ω 2 (ω > 0). In this case, the general solution of (1.13.24) is T (t) = A cos(ωct) + B sin(ωct).
(1.13.25)
The solutions of the reduced wave equation (1.13.23) can be obtained by further separation of variables. We illustrate the method for equation (1.13.16) which takes the form
1.13 Energy Integrals and Higher Dimensional Equations
where
105
1 ∂ 2 ∂v r + Δ v + ω2 v = 0, 3 r2 ∂r ∂r
(1.13.26)
1 ∂ ∂v 1 ∂2v . sin φ + Δ3 v ≡ sin φ ∂φ ∂φ sin2 φ ∂θ2
(1.13.27)
We seek separable solutions in the form v(r, θ, φ) = R(r)Y (θ, φ).
(1.13.28)
Substituting (1.13.28) into (1.13.26) gives 1 d 2 dR r2 dr (r dr ) + R/r2
ω2 R
=−
Δ3 Y = μ, Y
(1.13.29)
where μ is a separation constant, and the above equation becomes Δ3 Y + μY = 0, r 2 R + 2rR + ω 2 r2 − μ R = 0.
(1.13.30) (1.13.31)
The partial differential equation (1.13.30) is considerably more difficult to solve. It is useful to consider (θ, φ) as coordinates of a point on the surface of the unit sphere S(0, 1) with the center at the origin of R3 . Instead of finding all solutions of (1.13.30), it is useful to determine only those solutions Y (θ, φ) which are defined as C 2 functions on the whole of S(0, 1). Such solutions are periodic in θ with period 2π, and at the poles of the sphere (that is, at the points where φ = 0 and φ = π) the solutions tend to limits independent of θ. It can be shown (see Courant and Hilbert 1953, Vol. I, Chapter VII, §5) that equation (1.13.30) has nontrivial smooth solutions satisfying these conditions only when μ is equal to one of the integral values μm = m(m + 1), m = 0, 1, 2, . . . . For each such value of μm , there are (2m + 1) linearly independent solutions of (1.13.30) denoted by Ym(k) (θ, φ),
k = 1, 2, 3, . . . , (2m + 1).
These solutions are known as the Laplace spherical harmonics. For each μm , the radial equation (1.13.31) becomes (1.13.32) r2 R + 2rR + ω 2 r 2 − m(m + 1) R = 0. √ In terms of the new dependent variable w = rR, equation (1.13.32) can be rewritten as 2
1 w = 0. (1.13.33) r2 w + rw + ω 2 r2 − m + 2 This represents Bessel’s equation of the first kind of the order (m + 12 ) with ω as a parameter, and has two linearly independent Bessel’s functions solutions
106
1 Linear Partial Differential Equations
Jm+ 12 (ωr) and J−(m+ 12 ) (ωr),
(1.13.34)
where Jν (x) =
∞
(−1)n
n=0
( x2 )ν+2n , n!Γ (ν + n + 1)
(1.13.35)
with Γ representing the gamma function. The solutions (1.13.34) can be distin1 guished by their behavior at the origin, Jm+ 12 (ωr) behaves like rm+ 2 near r = 0, whereas J−(m+ 12 ) (ωr) behaves like r−(m+ 2 ) near r = 0. Consequently, for each m = 0, 1, 2, . . . , the radial equation (1.13.33) has two linearly independent solutions, 1
1 √ Jm+ 1 (ωr) 2 r
and
1 √ J−(m+ 1 ) (ωr). 2 r
(1.13.36)
These solutions behave like rm and r−m−1 , respectively, in the vicinity of r = 0. Thus, the corresponding product solutions of the reduced wave equation (1.13.26) are given by 1 √ Jm+ 1 (ωr)Ym(k) (θ, φ) 2 r
and
1 √ J−(m+ 1 ) (ωr)Ym(k) (θ, φ), 2 r
(1.13.37)
where m = 0, 1, 2, 3, . . . , and k = 1, 2, 3, . . . , (2m + 1). Thus, the method of separation of variables leads to the following solutions for the three-dimensional wave equation: Jm+ 12 (ωr) 1 cos(ωct) (k) √ Ym (θ, φ) , (1.13.38) sin(ωct) r J−(m+ 12 ) (ωr) where m = 0, 1, 2, . . . , and k = 1, 2, . . . , (2m + 1). These solutions are oscillatory in nature because of the harmonic dependence on time t. On the other hand, nonoscillatory solutions of (1.13.16) correspond to the case λ = −ω 2 < 0 (ω > 0). Obviously, the two linearly independent solutions of the time equation (1.13.24) are exp(ωct) and exp(−ωct), and the equations corresponding to (1.13.26), (1.13.30), and (1.13.31) are found by replacing ω 2 by −ω 2 . The equation corresponding to (1.13.33) is 2
1 w = 0. r 2 w + rw − ω 2 r 2 + m + 2
(1.13.39)
This represents Bessel’s equation with purely imaginary argument and has two linearly independent Bessel’s functions solutions, Im+ 12 (ωr)
and
I−(m+ 12 ) (ωr),
(1.13.40)
where Iν (x) is the modified Bessel function of the first kind of order ν defined by
1.13 Energy Integrals and Higher Dimensional Equations
Iν (x) =
∞
( x2 )ν+2n . n!Γ (ν + n + 1) n=0
107
(1.13.41)
We next consider the one-dimensional diffusion equation (1.6.36)–(1.6.38) to establish an energy inequality. Multiplying the diffusion equation (1.6.36) by u(x, t) and integrating with respect to x over [0, ] gives (uut ) dx = κ (uuxx ) dx. 0
0
Integrating the right-hand side by parts yields 1 2 dx = [κuux ]0 − κ u2x (x, t) dx u 2 0 0 t = −κ u2x (x, t) dx, by (1.6.37ab).
(1.13.42)
0
For mathematical reasons, we introduce another quantity E(t) by 1 2 E(t) = u dx, 2 0 which may be called the energy at time t. Evidently, for κ > 0, dE = −κ u2x (x, t) dx ≤ 0. dt 0
(1.13.43)
(1.13.44)
This result governing the energy is essentially a form of the entropy principle of thermodynamics. It states that the energy E(t) is a decreasing function of time t. Finally, integrating (1.13.44) with respect to time t over (t0 , t) gives t E(t) − E(t0 ) = −κ u2x (x, t) dx dt ≤ 0. 0
This means that
t0
u2 (x, t) dx ≤ 0
u2 (x, t0 ) dx.
(1.13.45)
0
This may be referred to as the energy inequality. Theorem 1.13.1 (Uniqueness). There exists a unique solution of the diffusion equation system (1.6.36)–(1.6.38). Proof. Suppose that there are two distinct solutions u(x, t) and v(x, t) of the diffusion equation system and w(x, t) = u(x, t) − v(x, t). It is easy to check that w(x, t) satisfies the diffusion equation (1.6.36) with w(x, 0) = 0 and w(0, t) = 0 = w(, t). It then follows from (1.13.44) that E(t) ≤ 0. But, by definition, E(t) ≥ 0. Evidently, E(t) ≡ 0 for t ≥ 0. This means that w(x, t) ≡ 0 for 0 ≤ x ≤ and t ≥ 0. Thus, u(x, t) = v(x, t). This proves the theorem.
108
1 Linear Partial Differential Equations
It follows from the derivation of the diffusion equation that heat is conducted away from regions of high temperature. This suggests that a temperature attains its maximum only initially or on boundaries. This is the essence and the origin of the maximum principle, which states that the maximum value of the temperature occurs either initially or on the boundaries. We now prove the maximum principle for the one-dimensional diffusion equation with the initial and boundary conditions u(x, 0) = f (x),
u(0, t) = f1 (t),
u(, t) = f2 (t),
and prove that a maximum of u(x, t) cannot occur at an interior point (x0 , t0 ) of the domain where 0 < x0 < , 0 < t0 < T . Theorem 1.13.2 (Maximum–Minimum Principle). If u(x, t) is continuous on the closed rectangle D = {0 ≤ x ≤ , 0 ≤ t ≤ T } and satisfies the diffusion equation (1.6.36) in D, then u(x, t) attains its maximum and minimum values on the lower base t = 0 or on the vertical sides x = 0, x = of D. Proof. We first prove the maximum principle. The proof is by contradiction. Let M be the maximum value of u in D. Contrary to the assertion of the theorem, assume that the maximum value of u on the lower base and the vertical sides of D is M − ε, where ε > 0. Suppose that u attains its maximum at a point (x0 , t0 ) in D so that u(x0 , t0 ) = M . We must have 0 < x0 < and t0 > 0. Consider the auxiliary function v(x, t) = u(x, t) +
ε (x − x0 )2 . 42
On the lower base and vertical sides of D, v(x, t) ≤ M − ε +
3ε ε =M− , 4 4
where v(x0 , t0 ) = M . Thus, the maximum value of v in D is not attained on the lower base and vertical sides of D. Let (x1 , t1 ) be a point where v attains its maximum. We must have 0 < x1 < and 0 < t1 < T . At (x1 , t1 ), v must satisfy the necessary condition for a maximum, that is, vt = 0 if t1 < T or vt = 0 if t1 = T , and vxx ≤ 0. Hence, at (x1 , t1 ), vt − vxx ≥ 0. On the other hand, vt − vxx = ut − uxx −
ε < 0. 22
This is a contradiction and the maximum part is proved. The minimum assertion becomes the maximum assertion when w = −u and w attains a maximum where u has a minimum. Since w satisfies all assumptions of the theorem, it must attain its maximum value on the lower base and vertical sides of D. Consequently, u must attain its minimum value there.
1.13 Energy Integrals and Higher Dimensional Equations
109
Finally, we consider the n-dimensional diffusion equation system ut = κ∇2n u, x ∈ Rn , t > 0, u(x, 0) = f (x), u(x, t) → 0 as |x| → ∞,
(1.13.46) (1.13.47)
where ∇2n ≡
∂2 ∂2 ∂2 2 + 2 + ··· + ∂x1 ∂x2 ∂x2n
(1.13.48)
is the n-dimensional Laplacian and x = (x1 , x2 , . . . , xn ) ∈ Rn
and
|x| =
& x21 + x22 + · · · + x2n .
Application of the n-dimensional Fourier transform gives the solution ∞ |x − ξ |2 1 ξ f (ξ ) exp − dξξ , (1.13.49) u(x, t) = 4κt (4πκt)n/2 −∞ where ξ = (ξ1 , ξ2 , . . . , ξn ). Example 1.13.1 (Transverse Vibration of a Thin Elastic Circular Membrane). The transverse vibration of a thin elastic circular membrane of radius a stretched over the rim is governed by the two-dimensional wave equation utt = c2 (uxx + uyy ),
(x, y) ∈ R2 , t > 0,
where c2 is a constant. In cylindrical polar coordinates, equation (1.13.50) is given by 1 1 2 utt = c urr + ur + 2 uθθ , 0 < r < a, 0 < θ ≤ 2π, t > 0. r r
(1.13.50)
(1.13.51)
For the membrane, the displacement vanishes along the rim, and hence, the boundary condition is u(a, θ) = 0.
(1.13.52)
The initial conditions on the displacement and velocity are u(r, θ, 0) = f (r, θ),
ut (r, θ, 0) = g(r, θ).
(1.13.53)
Using the separation of variables width u(r, θ, t) = R(r)Θ(θ)T (t) in (1.13.51) gives R R 1 ΘT + 2 Θ T, RΘT = R + (1.13.54) c2 r r where the primes denote differentiation with respect to the argument. Dividing by RΘT yields
110
1 Linear Partial Differential Equations
1 T Θ 1 1 + 2 = −k 2 , R = R + 2 c T R r r Θ
(1.13.55)
where −k 2 is a separation constant. Thus, equation (1.13.55) leads to T + c2 k 2 T = 0, 1 1 Θ R + R + 2 = −k 2 . R r r Θ Equation (1.13.57) can be further written as r2 1 Θ R + R + k2 r2 = − = α2 , R r Θ where α2 is a another separation constant. Consequently, we obtain r2 R + rR + k2 r 2 − α2 R = 0,
2
Θ + α Θ = 0.
(1.13.56) (1.13.57)
(1.13.58)
(1.13.59) (1.13.60)
The general solutions of (1.13.56) and (1.13.60) are T (t) = A cos(kct) + B sin(kct),
Θ (θ) = C cos αθ + D sin αθ,
(1.13.61) (1.13.62)
where A, B, C, D are constants of integration. Since the physical domain covers the entire circle, Θ and Θ must be periodic in θ, and hence α must be an integer n so that Θ(θ) is a linear combination of cos nθ and sin nθ. The radial equation (1.13.59) becomes (1.13.63) r 2 R + rR + k2 r2 − n2 R = 0. If the displacement depends only on the radial coordinate r from the center of the membrane and time t, then θ-dependence is absent, and hence, α = n = 0 and (1.13.63) reduces to the form rR + R + k 2 rR = 0,
0 ≤ r < a,
(1.13.64)
with the boundary condition R(a) = 0, |R(0)| < ∞. The general solution of (1.13.64) is R(r) = C1 J0 (kr) + C2 Y0 (kr),
(1.13.65)
where J0 and Y0 are Bessel functions of the first and second kinds of order zero. The condition |R(0)| < ∞ requires C2 = 0. Since Y0 becomes unbounded at r = 0, the boundary condition R(a) = 0 leads to the eigenvalue equation
1.13 Energy Integrals and Higher Dimensional Equations
C1 = 0,
J0 (ak) = 0,
111
(1.13.66)
which has an infinite number of positive roots at akn = λn , n = 1, 2, 3, . . . , so that the eigenvalues are given by 2 λn , (n = 1, 2, 3, . . .). (1.13.67) kn2 = a The corresponding eigenfunctions are
Rn (r) = C1n J0
λn r , a
(1.13.68)
with the eigenvalues kn2 . The solutions are
un (r, t) = an cos(ckn t) + bn sin(ckn t) J0
λn r , a
(1.13.69)
where an and bn are arbitrary constants. These solutions represent standing waves with fixed shape Jn (kn r) and amplitude Tn (t). Thus, these waves are similar to those observed in the vibrating string problem where the zeros of the sine functions are evenly spaced. However, the zeros of the Bessel functions are not evenly spaced, and hence, the sound emitted from a drum is quite different from that of stringed musical instruments, where the frequencies are integral multiples of the fundamental frequency, and zeros of the solution are equally spaced. Indeed, the musical quality of sound produced by stringed instruments is much more melodious than that of drums. According to the principle of superposition, the formal series solution is u(r, t) =
∞
(an cos ckn t + bn sin ckn t)Jn (rkn ),
(1.13.70)
n=1
where constants an and bn are determined by the initial conditions f (r) = u(r, 0) =
∞
an Jn (rkn ),
n=1 ∞
g(r) = ut (r, 0) =
ckn bn J0 (rkn ).
(1.13.71) (1.13.72)
n=1
These are called the Fourier–Bessel series. In particular, if g(r) = 0, (1.13.72) is satisfied when bn = 0, n = 1, 2, 3, . . . . In this case, the coefficients an are given by the formula (1.12.31), that is, a rf (r)J0 (rkn ) dr f (r), un (r) , (1.13.73) = 0 a 2 an = un 2 rJ0 (rkn ) dr 0 a 2 = 2 2 rf (r)J0 (rkn ) dr. (1.13.74) a J1 (akn ) 0 Thus, the problem is completely solved.
112
1 Linear Partial Differential Equations
Example 1.13.2 (Diffusion Equation in a Wedge-Shaped Region). The diffusion equation in a wedged-shaped region is 1 1 (1.13.75) ut = κ urr + ur + 2 uθθ , 0 ≤ r ≤ a, 0 ≤ θ ≤ α, r r where 0 ≤ α ≤ 2π. With the insulated boundary, the normal derivative of u on the boundary must vanish, that is, ur (a, θ) = uθ (r, 0) = uθ (r, α) = 0.
(1.13.76)
The initial condition is u(r, θ, 0) = f (r, θ).
(1.13.77)
Using the separation of variables, we write u(r, θ, t) = R(r)Θ(θ)T (t) so that (1.13.75) leads to the following equations: r2 R + rR + k2 r2 − ν 2 R = 0, R (a) = 0, (1.13.78) Θ + ν 2 Θ = 0, T (t) + κk 2 T (t) = 0,
Θ (0) = Θ (α) = 0, t > 0.
(1.13.79) (1.13.80)
We also assume that R(r) is bounded at r = 0. The equation (1.13.79) with Θ (0) = 0 implies that Θ(θ) = A cos(νθ),
(1.13.81)
. This leads to the Fourier series and the boundary condition at θ = α yields ν = nπ α in θ, Bessel functions of order ν in r, and exponential functions in t. Consequently, the eigenvalues for the Sturm–Liouville problem in r are k2 =
λmn a
2
and the corresponding eigenfunctions are λmn r , Rm (r) = Bm Jν a
,
(1.13.82)
m = 1, 2, 3, . . . ,
(1.13.83)
where λmn are the positive roots of Jν (x) = 0. For n = 0, we include the constant eigenfunction 1 corresponding to the eigenvalue k = 0 in the solution. Consequently, the general solution is u(r, θ, t) = a00 +
∞ ∞ n=0 m=1
anm Jν
λmn r a
nπθ κt cos exp − 2 λ2mn . α a (1.13.84)
1.13 Energy Integrals and Higher Dimensional Equations
The initial condition (1.13.77) leads to the result ∞ ∞ rλmn nπθ cos . f (r, θ) = a00 + anm Jν a α n=0 m=1
113
(1.13.85)
This Fourier–Bessel double series expansion allows us to find the coefficients a α 2 f (r, θ)r dr dθ, (1.13.86) a00 = αa2 0 0 a α 2λ2m0 rλm0 f (r, θ)J0 a0m = r dr dθ, (1.13.87) αa2 λ2m0 J02 (λm0 ) 0 0 a and for n, m ≥ 1 anm =
4λ2nm (λ2nm − ν 2 )−1 αa2 Jν2 (λnm )
a
α
f (r, θ)Jν 0
0
In particular, we consider the case of α =
π 2
rλnm a
nπθ cos r dr dθ. α (1.13.88)
and the initial temperature
u(r, θ, 0) = f (r, θ) = r2 cos 2θ. Consequently, 2
r cos 2θ = a00 +
∞ ∞ n=0 m=1
anm Jν
λnm r a
cos 2nθ,
(1.13.89)
where ν = nπ α = 2n. The left-hand side of (1.13.89) involves cos 2nθ only for n = 1. Clearly, the orthogonality relations for cos 2nθ in [0, π2 ] suggest that only the terms with n = 1 on the right-hand side will be nonzero. We cancel the cos 2nθ from both sides of (1.13.89) to obtain ∞ λm r 2 am J2 r = , (1.13.90) a m=1 where a1m = am and λm1 = λm . It follows from (1.13.90) that a λm r 2λ2m 3 am = 2 2 dr, r J 2 a (λm − 4)J22 (λm ) 0 a
(1.13.91)
where the integral on the right-hand side can be exactly evaluated so that 4 λm 4 a a a λm r 3 3 r J2 x J2 (x) dx = λ3m J3 (λm ). dr = a λm λm 0 0 (1.13.92) Consequently, the final solution is λm r κλ2m t λm J3 (λm ) . u(r, θ, t) = 2a cos 2θ exp − 2 J2 (λ2m − 4)J22 (λm ) a a m=1 2
∞
(1.13.93)
114
1 Linear Partial Differential Equations
1.14 Fractional Partial Differential Equations (a) Fractional Diffusion Equation The fractional diffusion equation is given by ∂αu ∂2u = κ , ∂tα ∂x2
x ∈ R, t > 0,
(1.14.1)
with the boundary and initial conditions u(x, t) → 0 as |x| → ∞, α−1 D u(x, t) t=0 = f (x) for x ∈ R, 0 t
(1.14.2) (1.14.3)
where κ is a diffusivity constant and 0 < α ≤ 1. Application of the Fourier transform to (1.14.1) with respect to x and using the boundary condition (1.14.2) yield
α ˜(k, t) 0 Dt u α−1 D u ˜(k, t) t=0 0 t
= −κk 2 u ˜, ˜ = f (k),
(1.14.4) (1.14.5)
where u ˜(k, t) is the Fourier transform of u(x, t) defined by (1.7.1). The Laplace transform solution of (1.14.4) and (1.14.5) yields ˜ u ¯(k, s) =
f˜(k) . (sα + κk 2 )
(1.14.6)
The inverse Laplace transform of (1.14.6) gives u ˜(k, t) = f˜(k)tα−1 Eα,α −κk 2 tα ,
(1.14.7)
where Eα,β is the Mittag-Leffler function defined by Eα,β (z) =
∞
zm , Γ (αm + β) m=0
α > 0, β > 0.
Finally, the inverse Fourier transform leads to the solution ∞ G(x − ξ, t)f (ξ) dξ, u(x, t) =
(1.14.8)
(1.14.9)
−∞
where 1 G(x, t) = π
∞
−∞
tα−1 Eα,α −κk2 tα cos kx dk.
(1.14.10)
This integral can be evaluated by using the Laplace transform of G(x, t) as
1.14 Fractional Partial Differential Equations
G(x, s) =
1 π
∞
−∞
cos kx dk 1 −α/2 |x| α/2 √ √ s , s = exp − sα + κk2 κ 4κ
115
(1.14.11)
where (m) L tmα+β−1 Eα,β ±atα =
m!sα−β , (sα ∓ a)m+1
(1.14.12)
and (m)
Eα,β (z) =
dm Eα,β (z). dz m
(1.14.13)
The inverse Laplace transform of (1.14.11) gives the explicit solution 1 α −1 α α 2 √ G(x, t) = , (1.14.14) t W −ξ, − , 2 2 4κ |x| where ξ = √κt α/2 , and W (z, α, β) is the Wright function (see Erdélyi 1955, formula 18.1 (27)) defined by
W (z, α, β) =
∞
zn . n!Γ (αn + β) n=0
(1.14.15)
It is important to note that when α = 1, the initial-value problem (1.14.1)– (1.14.3) reduces to the classical diffusion problem, and solution (1.14.9) reduces to the classical solution (1.7.63) because 1 1 1 x x2 1 G(x, t) = √ W −√ , − , exp − =√ . (1.14.16) 4κt κt 2 2 4κt 4πκt The fractional diffusion equation (1.14.1) has also been solved by other authors including Schneider and Wyss (1989), Mainardi (1994, 1995), Debnath (2003a, 2003b), and Nigmatullin (1986) with a physical realistic initial condition u(x, 0) = f (x),
x ∈ R.
(1.14.17)
The solutions obtained by these authors are in total agreement with (1.14.9). It is noted that the order α of the derivative with respect to time t in equation (1.14.1) can be of arbitrary real order including α = 2 so that it may be called the fractional diffusion–wave equation. For α = 2, it becomes the classical wave equation. Equation (1.14.1) with 1 < α ≤ 2 will be solved next in some detail. (b) Fractional Nonhomogeneous Wave Equation The fractional nonhomogeneous wave equation is given by 2 ∂αu 2∂ u − c = q(x, t), ∂tα ∂x2
x ∈ R, t > 0
(1.14.18)
116
1 Linear Partial Differential Equations
with the initial condition u(x, 0) = f (x),
ut (x, 0) = g(x),
x ∈ R,
(1.14.19)
where c is a constant and 1 < α ≤ 2. Application of the joint Laplace transform with respect to t and the Fourier transform with respect to x gives the transform solution ˜¯(k, s) = u
f˜(k)sα−1 g˜(k)sα−2 q˜¯(k, s) + + α , sα + c2 k 2 sα + c2 k 2 s + c2 k 2
(1.14.20)
where k is the Fourier transform variable and s is the Laplace transform variable. The inverse Laplace transform produces the following result: sα−1 sα−2 −1 + g ˜ (k)L u ˜(k, t) = f˜(k)L−1 α 2 2 sα + c2 k 2 s + c k q˜ ¯(k, s) , (1.14.21) + L−1 α s + c2 k 2 which, by (1.14.12), is
= f˜(k)Eα,1 −c2 k 2 tα + g˜(k)tEα,2 −c2 k 2 tα t + q˜(k, t − τ )τ α−1 Eα,α −c2 k 2 τ α dτ.
(1.14.22)
0
Finally, the inverse Fourier transform gives the formal solution ∞ 1 u(x, t) = √ f˜(k)Eα,1 −c2 k 2 tα eikx dk 2π −∞ ∞ 1 t˜ g (k)Eα,2 −c2 k 2 τ α eikx dk +√ 2π −∞ ∞ t 1 α−1 τ dτ q˜(k, t − τ )Eα,α −c2 k 2 τ α eikx dk. +√ 2π 0 −∞ (1.14.23) In particular, when α = 2, the fractional wave equation (1.14.18) reduces to the classical wave equation (1.9.10). In this particular case, we use E2,1 −c2 k 2 t2 = cosh(ickt) = cos(ckt), (1.14.24) 2 2 2 1 sinh(ickt) = sin ckt. (1.14.25) tE2,2 −c k t = t · ickt ck Consequently, solution (1.14.23) reduces to solution (1.9.15) for α = 2 ∞ ∞ 1 sin(ckt) ikx 1 g˜(k) f˜(k) cos(ckt)eikx dk + √ u(x, t) = √ e dk ck 2π −∞ 2π −∞ t ∞ 1 sin ck(t − τ ) ikx +√ e dk dτ q˜(k, τ ) (1.14.26) k 2πc 0 −∞
1.14 Fractional Partial Differential Equations
=
x+ct 1 1 f (x − ct) + f (x + ct) + g(ξ) dξ 2 2c x−ct x+c(t−τ ) 1 t dτ q(ξ, τ ) dξ. + 2c 0 x−c(t−τ )
117
(1.14.27)
We now derive the solution of the inhomogeneous fractional diffusion equation (1.14.18) with c2 = κ and g(x) = 0. In this case, the joint transform solution (1.14.20) becomes ˜ u ¯(k, s) =
f˜(k)sα−1 q˜ ¯(k, s) + α α 2 (s + κk ) (s + κk 2 )
(1.14.28)
which is inverted by (1.14.12) to obtain u ˜(k, t) in the form = f˜(k)Eα,1 −κk 2 tα +
t
(t − τ )α−1 Eα,α −κk 2 (t − τ )α q˜(k, τ ) dτ.
0
(1.14.29) Finally, the inverse Fourier transform gives the exact solution ∞ 1 √ f˜(k)Eα,1 −κk2 tα eikx dk u(x, t) = 2π −∞ t ∞ 1 +√ dτ (t − τ )α−1 Eα,α −κk 2 (t − τ )α q˜(k, τ )eikx dk. 2π 0 −∞ (1.14.30) Application of the convolution theorem of the Fourier transform gives the final solution in the form ∞ G1 (x − ξ, t)f (ξ) dξ u(x, t) = −∞
t
(t − τ )α−1 dτ
+ 0
∞ −∞
G2 (x − ξ, t − τ )q(ξ, τ ) dξ,
(1.14.31)
where G1 (x, t) =
1 2π
∞ −∞
eikx Eα,1 −κk 2 tα dk
(1.14.32)
eikx Eα,α −κk 2 tα dk.
(1.14.33)
and 1 G2 (x, t) = 2π
∞ −∞
In particular, when α = 1, the classical solution of the nonhomogeneous diffusion equation (1.14.18) is obtained in the form
118
1 Linear Partial Differential Equations
∞
u(x, t) = −∞
G1 (x − ξ, t)f (ξ) dξ
t
+
∞
dτ 0
−∞
G2 (x − ξ, t − τ )q(ξ, τ ) dξ,
(1.14.34)
x2 1 exp − . 4κt 4πκt
(1.14.35)
where G1 (x, t) = G2 (x, t) = √
In the case of classical homogeneous diffusion equation (1.14.18), solutions (1.14.30) and (1.14.34) are in perfect agreement with those of Mainardi (1996a, 1996b), who obtained the solution by using the Laplace transform method together with complicated evaluation of the Laplace inversion integral and the auxiliary function M (z, α). However, Mainardi (1996a, 1996b) obtained the solution in terms of M (z, α2 ) and discussed the nature of the solution for different values of α. He made some comparison between the ordinary diffusion (α = 1) and fractional diffusion (α = 12 and α = 23 ). For cases α = 43 and α = 32 , the solution exhibits a striking difference from ordinary diffusion with a transition from the Gaussian function centered at z = 0 (ordinary diffusion) to the Dirac delta function centered at z = 1 (wave propagation). This indicates a possibility of an intermediate process between diffusion and wave propagation. A special difference is observed between the solutions of the fractional diffusion equation (0 < α ≤ 1) and the fractional wave equation (1 < α ≤ 2). In addition, the solution exhibits a slow process for the case with 0 < α ≤ 1 and an intermediate process for 1 < α ≤ 2. (c) Fractional-Order Diffusion Equation in Semi-Infinite Medium We consider the fractional-order diffusion equation in a semi-infinite medium x > 0, when the boundary is kept at a temperature u0 f (t) and the initial temperature is zero in the whole medium. Thus, the initial boundary-value problem is governed by the equation ∂αu ∂ 2u = κ 2, α ∂t ∂x
0 < x < ∞, t > 0,
(1.14.36)
with u(x, t = 0) = 0, x > 0, u(x = 0, t) = u0 f (t), t > 0 and u(x, t) → 0 as x → ∞.
(1.14.37) (1.14.38)
Application of the Laplace transform with respect to t gives α d2 u ¯ s u ¯(x, s) = 0, x > 0, − dx2 κ u ¯(x = 0, s) = u0 f¯(s), u ¯(x, s) → 0 as x → ∞.
(1.14.39)
Evidently, the solution of this transformed boundary-value problem is
(1.14.40)
1.14 Fractional Partial Differential Equations
u ¯(x, s) = u0 f¯(s) exp(−ax),
119
(1.14.41)
1
where a = (sα /κ) 2 . Thus, the solution is given by
t
f (t − τ )g(x, τ ) dτ = u0 f (t) ∗ g(x, t),
u(x, t) = u0
(1.14.42)
0
where g(x, t) = L−1 exp(−ax) . In this case, α = 1 and f (t) = 1, and the solution (1.14.41) becomes u0 s u ¯(x, s) = exp −x , (1.14.43) s κ which yields the classical solution in terms of the complementary error function (see Debnath 1995) x u(x, t) = u0 erfc √ . (1.14.44) 2 κt In the classical case (α = 1), the more general solution is given by
t
f (t − τ )g(x, τ ) dτ = u0 f (t) ∗ g(x, t),
u(x, t) = u0
(1.14.45)
0
where x2 s x exp − g(x, t) = L−1 exp −x = √ . κ 4κt 2 πκt3
(1.14.46)
(d) The Fractional Stokes and Rayleigh Problems in Fluid Dynamics The classical Stokes problem (see Debnath 1995) deals with the unsteady boundary layer flows induced in a semi-infinite viscous fluid bounded by an infinite horizontal disk at z = 0 due to nontorsional oscillations of the disk in its own plane with a given frequency ω. When ω = 0, the Stokes problem reduces to the classical Rayleigh problem where the unsteady boundary layer flow is generated in the fluid from rest by moving the disk impulsively in its own plane with constant velocity U . We consider the unsteady fractional boundary layer equation for the fluid velocity u(z, t) ∂ 2u ∂αu = ν 2, α ∂t ∂z
0 < z < ∞, t > 0,
(1.14.47)
with the given boundary and initial conditions u(0, t) = U f (t),
u(z, t) → 0
as z → ∞, t > 0,
(1.14.48)
120
1 Linear Partial Differential Equations
u(z, 0) = 0 for all z > 0,
(1.14.49)
where ν is the kinematic viscosity, U is a constant velocity, and f (t) is an arbitrary function of time t. Application of the Laplace transform with respect to t gives ¯ d2 u , 0 < z < ∞, dz 2 u ¯(0, s) = U f¯(s), u ¯(z, s) → 0 as z → ∞.
sα u ¯(z, s) = ν
(1.14.50) (1.14.51)
Using the Fourier sine transform (see Debnath 1995) with respect to z yields U s (k, s) =
2 k f¯(s) νU . α π (s + νk 2 )
(1.14.52)
The inverse Fourier sine transform of (1.14.52) leads to the solution u ¯(z, s) =
∞ 2 k sin kz ¯ νU f (s) dk, α + νk 2 ) π (s 0
(1.14.53)
and the inverse Laplace transform gives the solution for the velocity u(z, t) =
2 νU π
∞
t
k sin kz dk 0
f (t − τ )τ α−1 Eα,α −νk2 τ α dτ.
0
(1.14.54) When f (t) = exp(iωt), the solution of the fractional Stokes problem is u(z, t) =
t 2νU iωt ∞ k sin kz dk e−iωτ τ α−1 Eα,α −νk2 τ α dτ. e π 0 0 (1.14.55)
When α = 1, solution (1.14.55) reduces to the classical Stokes solution u(z, t) =
2νU π
∞
1 − e−νtk
0
2
k sin kz dk. (iω + νk 2 )
(1.14.56)
For the fractional Rayleigh problem, f (t) = 1 and the solution follows from (1.14.54) in the form u(z, t) =
2νU π
∞
t
k sin kz dk 0
τ α−1 Eα,α −νk 2 τ α dτ.
0
This solution reduces to the classical Rayleigh solution when α = 1 as
(1.14.57)
1.14 Fractional Partial Differential Equations
u(z, t) =
2νU π
∞
t
k sin kz dk 0
121
E1,1 −ντ k2 dτ
0
∞ t 2νU = k sin kz dk exp −ντ k 2 dτ π 0 0 ∞ 2U −νtk2 sin kz dk, 1−e = π k 0 which, by (2.10.10) of Debnath (1995),
2U π π z z = − erf √ = U erfc √ . π 2 2 2 νt 2 νt
(1.14.58)
The above analysis is in full agreement with the classical solutions of the Stokes and Rayleigh problems (see Debnath 1995). (e) The Fractional Unsteady Couette Flow We consider the unsteady viscous fluid flow between the plate at z = 0 at rest and the plate z = h in motion parallel to itself with a variable velocity U (t) in the x-direction. The fluid velocity u(z, t) satisfies the fractional equation of motion ∂2u ∂α u = P (t) + ν 2 , α ∂t ∂z
0 ≤ z ≤ h, t > 0,
(1.14.59)
with the boundary and initial conditions u(0, t) = 0 and u(h, t) = U (t), t > 0, u(z, t) = 0 at t ≤ 0 for 0 ≤ z ≤ h,
(1.14.60) (1.14.61)
where − ρ1 px = P (t) and ν is the kinematic viscosity of the fluid. We apply the joint Laplace transform with respect to t and the finite Fourier sine transform with respect to z defined by ∞ h nπz ¯˜s (n, s) = dz (1.14.62) e−st dt u(z, t) sin u h 0 0 to the system (1.14.59)–(1.14.61) so that the transform solution is 1 n n+1 U (s) ¯˜s (n, s) = P (s) a [1 − (−1) ] + νa(−1) u , α 2 α (s + νa ) (s + νa2 )
where a = ( nπ h ) and n is the finite Fourier sine transform variable. Thus, the inverse Laplace transform yields t 1 n P (t − τ )τ α−1 Eα,α −νa2 τ α dτ u ˜s (n, t) = 1 − (−1) a 0 t U (t − τ )τ α−1 Eα,α −νa2 τ α dτ. + νa(−1)n+1 0
(1.14.63)
(1.14.64)
122
1 Linear Partial Differential Equations
Finally, the inverse finite Fourier sine transform leads to the solution u(z, t) =
∞ nπz 2 u ˜s (n, t) sin . h n=1 h
(1.14.65)
If, in particular, when α = 1, P (t) = const., and U (t) = const., then solution (1.14.65) reduces to the solution of the generalized Couette flow (see p. 277 Debnath 1995). (f) Fractional Axisymmetric Wave–Diffusion Equation The fractional axisymmetric equation in an infinite domain 2 ∂ u 1 ∂u ∂αu = a + , 0 < r < ∞, t > 0, ∂tα ∂r2 r ∂r
(1.14.66)
is called the diffusion or wave equation accordingly as a = κ or a = c2 . For the fractional diffusion equation, we prescribe the initial condition u(r, 0) = f (r),
0 < r < ∞.
(1.14.67)
Application of the joint Laplace transform with respect to t and the Hankel transform of zero order (see Debnath 1995) with respect to r to (1.14.66), (1.14.67) gives the transform solution sα−1 f˜(k) ¯ , u ˜(k, s) = α (s + κk 2 ) where k, s are the Hankel and Laplace transform variables, respectively. The joint inverse transform leads to the solution ∞ u(r, t) = kJ0 (kr)f˜(k)Eα,1 −κk2 tα dk,
(1.14.68)
(1.14.69)
0
where J0 (kr) is the Bessel function of the first kind of order zero and f˜(k) is the Hankel transform of f (r). When α = 1, solution (1.14.69) reduces to the classical solution (1.10.22). On the other hand, we can solve the wave equation (1.14.66) with a = c2 and the initial conditions u(r, 0) = f (r),
ut (r, 0) = g(r)
for0 < r < ∞,
(1.14.70)
provided the Hankel transforms of f (r) and g(r) exist. Application of the joint Laplace and Hankel transform leads to the transform solution sα−1 f˜(k) sα−2 g˜(k) ¯ u ˜(k, s) = α + α . 2 2 (s + c k ) (s + c2 k 2 )
(1.14.71)
1.14 Fractional Partial Differential Equations
123
The joint inverse transformation gives the solution
∞
u(r, t) = 0
kJ0 (k, r)f˜(k)Eα,1 −c2 k 2 tα dk
∞
kJ0 (k, r)˜ g (k)tEα,2 −c2 k 2 tα dk.
+
(1.14.72)
0
When α = 2, (1.14.72) reduces to the classical solution (1.10.12). In a finite domain 0 ≤ r ≤ a, the fractional diffusion equation (1.14.66) can be solved by using the joint Laplace and finite Hankel transform with the boundary and initial data u(r, t) = f (t)
on r = a, t > 0,
u(r, 0) = 0 for all r in (0, a).
(1.14.73) (1.14.74)
Application of the joint Laplace and finite Hankel transform of zero order (see pp. 317, 318, Debnath 1995) yields the solution u(r, t) =
∞ 2 J0 (rki ) , u ˜(ki , t) 2 2 a i=1 J1 (aki )
(1.14.75)
where
t
u ˜(ki , t) = (aκki )J1 (aki )
f (t − τ )τ α−1 Eα,α −κki2 τ α dτ.
(1.14.76)
0
When α = 1, (1.14.75) reduces to (11.4.7) obtained by Debnath (1995). Similarly, the fractional wave equation (1.14.66) with a = c2 in a finite domain 0 ≤ r ≤ a with the boundary and initial conditions u(r, t) = 0
on r = a, t > 0,
u(r, 0) = f (r)
and
ut (r, 0) = g(r)
(1.14.77) for 0 < r < a,
(1.14.78)
can be solved by means of the joint Laplace and finite Hankel transforms. The solution of this problem is u(r, t) =
∞ J0 (rki ) 2 u ˜(ki , t) 2 , a2 i=1 J1 (aki )
(1.14.79)
where u ˜(ki , t) = f˜(ki )Eα,1 −c2 ki2 tα + g˜(ki )tEα,2 −c2 ki2 tα .
(1.14.80)
When α = 2, solution (1.14.79) reduces to the solution (11.4.26) obtained by Debnath (1995).
124
1 Linear Partial Differential Equations
(g) The Fractional Schrödinger Equation in Quantum Mechanics The one-dimensional fractional Schrödinger equation for a free particle of mass m (see (1.7.71)) is 2 ∂ 2 ψ ∂ αψ , −∞ < x < ∞, t > 0, (1.14.81) i α = − ∂t 2m ∂x2 ψ(x, 0) = ψ0 (x), −∞ < x < ∞, (1.14.82) ψ(x, t) → 0 as |x| → ∞, (1.14.83) where ψ(x, t) is the wave function, h = 2π = 6.625 × 10−27 erg sec = 4.14 × 10−21 MeV sec is the Planck constant, and ψ0 (x) is an arbitrary function. Application of the joint Laplace and Fourier transform to (1.14.81)–(1.14.83) gives the solution in the transform space in the form i sα−1 ψ˜0 (k) ˜ a= , (1.14.84) ψ(k, s) = α s + ak2 2m where k, s represent the Fourier and the Laplace transform variables. The use of the joint inverse transform yields the solution ∞ 1 √ ψ(x, t) = eikx ψ˜0 (k)Eα,1 −ak2 tα dk 2π −∞ = F −1 ψ˜0 (k)Eα,1 −ak 2 tα ,
(1.14.86)
which is, by the Convolution Theorem 1.7.1 of the Fourier transform, ∞ G(x − ξ, t)ψ0 (ξ) dξ, =
(1.14.87)
(1.14.85)
−∞
where
1 G(x, t) = √ F −1 Eα,1 −ak 2 tα 2π ∞ 1 eikx Eα,1 −ak 2 tα dk. = 2π −∞
When α = 1, solution (1.14.87) becomes ∞ ψ(x, t) = G(x − ξ, t)ψ0 (ξ) dξ,
(1.14.88)
(1.14.89)
−∞
where Green’s function G(x, t) is given by ∞ 1 G(x, t) = eikx E1,1 −ak 2 t dk 2π −∞ ∞ 1 exp ikx − atk 2 dk = 2π −∞ 1 x2 = √ exp − . 4at 4πat
(1.14.90)
This solution (1.14.89) is in perfect agreement with the classical solution obtained by Debnath (1995).
1.15 Exercises
125
1.15 Exercises 1. Classify each of the partial differential equations below as either hyperbolic, parabolic, or elliptic, determine the characteristics, and transform the equations to canonical form: (b) 2uxx − 3uxy + uyy = y, (a) 4uxx + 5uxy + uyy + ux + uy = 2, (d) uxx + yuyy = 0, (c) yuxx + (x + y)uxy + xuyy = 0, (e) yuxx − 2uxy + ex uyy + x2 ux − u = 0, (f) uxx + xuyy = 0, (h) 3yuxx − xuyy = 0, (g) x2 uxx + 4yuxy + uyy + 2ux = 0, (i) uxx + 2xuxy + a2 uyy + u = 5, (j) y 2 uxx + x2 uyy = 0. 2. Determine the nature of the following equations and reduce them to canonical form: (b) uxx − xuyy = 0, (a) x2 uxx + 4xyuxy + y 2 uyy = 0, (c) uxx − 2uxy + 3uyy + 24uy + 5u = 0, (d) uxx + sech4 xuyy = 0, (e) uxx + 6yuxy + 9y 2 uyy + 4u = 0, (f) uxx − sech4 xuyy = 0, 2 (g) uxx + 2cosecyuxy + cosec yuyy = 0, (h) uxx − 5uxy + 5uyy = 0. 3. For what values of m is uxx −mx uxy +4x2 uyy = 0 (a) hyperbolic, (b) parabolic, or (c) elliptic? For m = 0, reduce to canonical form. 4. (a) Show that the nonlinear equation u2 uxx + 2ux uy uxy − u2 uyy = 0 is hyperbolic for every solution u(x, y). (b) Show that the nonlinear equation for the velocity potential u(x, y) 1 − u2x uxx − 2ux uy uxy + 1 − u2y uyy = 0 in certain kinds of compressible fluid flow is (i) elliptic, (ii) parabolic, or (iii) hyperbolic for those solutions such that |∇u| < 1, |∇u| = 1, or |∇u| > 1. 5. Solve Example 1.6.2 with the boundary conditions ux (0, t) = 0 = ux (, t)
for t > 0,
leaving the initial condition (1.6.38) unchanged. 6. Use the separation of variables to solve the Laplace equation uxx + uyy = 0,
0 ≤ x ≤ a, 0 ≤ y ≤ b,
with u(0, y) = 0 = u(a, y) for 0 ≤ y ≤ b, and u(x, 0) = f (x) for 0 < x < a; u(x, b) = 0 for 0 ≤ x ≤ a. 7. Show that the eigenvalues of the eigenvalue problem utt + c2 uxxxx = 0, 0 < x < , t > 0, u(0, t) = 0 = u(, t) for t ≥ 0, ux (0, t) = 0 = ux (, t) for t ≥ 0,
126
1 Linear Partial Differential Equations
satisfy the equation cos(λ) cosh(λ) = 1. 8. Solve the problem in Exercise 4 with the boundary conditions u(x, 0) = f (x), ut (x, 0) = g(x) u(0, t) = 0 = u(, t) for t > 0,
for 0 ≤ x ≤ ,
for t > 0.
uxx (0, t) = 0 = uxx (, t)
9. Solve Example 1.6.2 with the initial data x if 0 ≤ x ≤ 2 , f (x) = − x if 2 ≤ x ≤ . 10. Solve Example 1.6.1 with the initial data hx if 0 ≤ x ≤ a, a (i) f (x) = h( − x)/( − a) if a ≤ x ≤ , u0 x (ii) f (x) = 0 and g(x) =
and g(x) = 0.
a
u0 ( − x)/( − a)
if 0 ≤ x ≤ a, if a ≤ x ≤ .
11. Solve the biharmonic wave equation utt + uxxxx = 0, −∞ < x < ∞, t > 0, u(x, 0) = f (x), ut (x, 0) = 0 for − ∞ < x < ∞. 12. Find the solution of the dissipative wave equation utt − c2 uxx + αut = 0,
−∞ < x < ∞, t > 0, ∂u u(x, 0) = f (x), = g(x) for − ∞ < x < ∞, ∂t t=0
where α > 0 is the dissipation parameter. 13. Solve the Cauchy problem for the linear Klein–Gordon equation utt − c2 uxx + a2 u = 0,
−∞ < x < ∞, t > 0, ∂u u(x, 0) = f (x), = g(x) for − ∞ < x < ∞. ∂t t=0
14. Solve the telegraph equation utt − c2 uxx + ut − aux = 0,
−∞ < x < ∞, t > 0, ∂u u(x, 0) = f (x), = g(x) for − ∞ < x < ∞. ∂t t=0
1.15 Exercises
127
Show that the solution is unstable when c2 < a2 . If c2 > a2 , show that the bounded integral solution is given by ∞ 1 A(k) exp −k 2 c2 − a2 t + ik(x + at) dk, u(x, t) = √ 2π −∞ where A(k) is given in terms of the transformed functions of the initial data. Hence, deduce the asymptotic solution, as t → ∞, in the form
(x + at)2 π exp − 2 . u(x, t) = A(0) 2(c2 − a2 )t 4(c − a2 )t 15. The transverse vibration of an infinite elastic beam of mass m per unit length and bending stiffness EI is governed by EI , −∞ < x < ∞, t > 0. m Solve this equation subject to the boundary and initial data utt + a2 uxxxx = 0,
a2 =
u(0, t) = 0 for all t > 0, u(x, 0) = φ(x), and ut (x, 0) = ψ (x)
for 0 < x < ∞.
Show that the Fourier transform solution is 1 U (k, t) = Φ(k) cos atk2 − Ψ (k) sin atk2 . a Find the integral solution for u(x, t). 16. Solve the Lamb (1904) problem in geophysics that satisfies the Helmholtz equation in an infinite elastic half-space uxx + uzz +
ω2 u = 0, c22
−∞ < x < ∞,
z > 0,
where ω is the frequency and c2 is the shear wave speed. At the surface of the half-space (z = 0), the boundary condition relating the surface stress to the impulsive point load distribution is given by ∂u = −P δ(x) at z = 0, ∂z where μ is one of the Lamé constants, P is a constant, and μ
u(x, z) → 0
as z → ∞ for − ∞ < x < ∞.
Show that the solution in terms of polar coordinates is P (2) ωr H u(x, z) = 2iμ 0 c2 12 2c2 πi iωr P − for ωr c2 . exp ∼ 2iμ πωr 4 c2
128
1 Linear Partial Differential Equations
17. Find the solution of the Cauchy–Poisson problem (Debnath 1994, p. 83) in inviscid water of infinite depth which is governed by −∞ < x < ∞, −∞ < z ≤ 0, t > 0,
φxx + φzz = 0, φz − ηt = 0,
on z = 0, t > 0,
φt + gη = 0
φz → 0 as z → −∞. φ(x, 0, 0) = 0, and η(x, 0) = P δ(x), where φ = φ(x, z, t) is the velocity potential, η(x, t) is the free surface elevation, and P is a constant. Derive the asymptotic solution for the free surface elevation as t → ∞. 18. Obtain the solutions for the velocity potential φ(x, z, t) and the free surface elevation η(x, t) involved in the two-dimensional surface waves in water of finite (or infinite) depth h. The governing equation, boundary, and free surface conditions and initial conditions (see Debnath 1994, p. 92) are −h ≤ z ≤ 0, −∞ < x < ∞, t > 0, p(x) exp(iωt), z = 0, t > 0,
φxx + φzz = 0, − Pρ
φt + gη = φ z − ηt = 0
φ(x, z, 0) = 0 = η(x, 0)
for all x and z.
19. Solve the steady-state surface wave problem (Debnath 1994, p. 47) on a running stream of infinite depth due to an external steady pressure applied to the free surface. The governing equation and the free surface conditions are φxx + φzz = 0,
−∞ < x < ∞, −∞ < z < 0, t > 0,
φx + U φx + gη = − Pρ δ(x) exp(εt), ηt + U ηx = φz φz → 0
on z = 0 (ε > 0),
as z → −∞.
where U is the stream velocity, φ(x, z, t) is the velocity potential, and η(x, t) is the free surface elevation. 20. Apply the Fourier transform to solve the initial-value problem for the dissipative wave equation utt = c2 uxx + αuxxt , −∞ < x < ∞, t > 0, u(x, 0) = f (x), ut (x, 0) = αf (x) for − ∞ < x < ∞, where α is a positive constant. 21. Use the Fourier transform to solve the boundary-value problem uxx + uyy = −x exp −x2 , −∞ < x < ∞, 0 < y < ∞,
1.15 Exercises
129
u(x, 0) = 0, for −∞ < x < ∞, u and its derivative vanish as y → ∞. Show that 2 ∞ sin kx k 1 exp − dk. 1 − exp(−ky) u(x, y) = √ k 4 4π 0 22. Solve the initial-value problem (Debnath 1994, p. 115) for the two-dimensional surface waves at the free surface of a running stream of velocity U . The problem satisfies the following equation, boundary, and initial conditions: φxx + φzz = 0, −∞ < x < ∞, −h ≤ z ≤ 0, t > 0, φx + U φx + gη = − Pρ δ(x) exp(iωt), on z = 0, t > 0, ηt + U ηx − φz = 0 φ(x, z, 0) = η(x, 0) = 0,
for all x and z.
23. Apply the Fourier transform to solve the equation uxxxx + uyy = 0,
−∞ < x < ∞, y ≥ 0,
satisfying the conditions u(x, 0) = f (x),
uy (x, 0) = 0 for − ∞ < x < ∞,
and u(x, y) and its partial derivatives vanish as |x| → ∞. 24. The transverse vibration of a thin membrane of great extent satisfies the wave equation c2 (uxx + uyy ) = utt ,
−∞ < x, y < ∞, t > 0,
with the initial and boundary conditions u(x, y, t) → 0 as |x| → ∞, |y| → ∞ for all t ≥ 0, u(x, y, 0) = f (x, y), ut (x, y, 0) = 0 for all x, y. Apply the double Fourier transform method to solve this problem. 25. Solve the diffusion problem with a source function q(x, t) ut = κuxx + q(x, t), −∞ < x < ∞, t > 0, u(x, 0) = 0 for − ∞ < x < ∞. Show that the solution is
t ∞ 1 1 (x − k)2 dk. u(x, t) = √ (t − τ )− 2 dτ q(k, τ ) exp − 4κ(t − τ ) 4πκ 0 −∞ 26. Apply the triple Fourier transform to solve the initial-value problem ut = κ(uxx + uyy + uzz ), u(x, 0) = f (x) for all x, y, z, where x = (x, y, z).
−∞ < x, y, z < ∞, t > 0,
130
1 Linear Partial Differential Equations
27. Use the double Fourier transform to solve the telegraph equation utt + aut + bu = c2 uxx , −∞ < x, t < ∞, u(0, t) = f (t), ux (0, t) = g(t), for − ∞ < t < ∞, where a, b, c are constants and f (t) and g(t) are arbitrary functions of t. 28. Use the Fourier transform to solve the Rossby wave problem in an inviscid βplane ocean bounded by walls at y = 0 and y = 1, where y and x represent vertical and horizontal directions. The fluid is initially at rest and then, at t = 0+, an arbitrary disturbance localized to the vicinity of x = 0 is applied to generate Rossby waves. This problem satisfies the Rossby wave equation ∂ 2 ∇ − κ2 ψ + βψx = 0, ∂t
−∞ < x < ∞, 0 ≤ y ≤ 1, t > 0,
with the boundary and initial conditions ψx (x, y) = 0 for 0 < x < ∞, y = 0 and y = 1, ψ(x, y, t) = ψ0 (x, y) at t = 0 for all x and y. 1 exp{ik0 x − ( xa )2 }. Examine the case for ψ0n (x) = α√ 2 29. The equations for the current I(x, t) and potential V (x, t) at a point x and time t of a transmission line containing resistance R, inductance L, capacitance C, and leakage inductance G are
LI t + RI = −Vx ,
and
CVt + GV = −Ix .
Show that both I and V satisfy the telegraph equation 1 utt − uxx + aut + bu = 0, c2 where c2 = (LC)−1 , a = LG + RC, and b = RG. Solve the telegraph equation for the following cases with R = 0 and G = 0: (a) V (x, t) = V0 H(t) at x = 0, t > 0, V (x, t) → 0 as x → ∞, t > 0, where V0 is constant. (b) V (x, t) = V0 cos ωt at x = 0, t > 0, V (x, t) → 0 as x → ∞, t > 0. 30. Solve the telegraph equation in Exercise 29 with V (x, 0) = 0 for (a) the Kelvin ideal cable line (L = 0 = G) with the boundary data V (0, t) = V0 = const., V (x, t) → 0 as x → ∞ for t > 0. (b) the noninductive leaky cable (L = 0) with the boundary conditions V (0, t) = H(t) and V (x, t) → 0 as x → ∞ for t > 0. 31. Solve the telegraph equation in Exercise 29 with V (x, 0) = 0 = Vt (x, 0) for G the Heaviside distortionless cable ( R L = C = const. = k) with the boundary data V (0, t) = V0 f (t) and V (x, t) → 0 as x → ∞ for t > 0, where V0 is constant and f (t) is an arbitrary function of t. Explain the physical significance of the solution.
1.15 Exercises
131
32. Solve the inhomogeneous partial differential equation uxt = −ω sin ωt, t > 0, u(x, 0) = x, u(0, t) = 0. 33. Find the solution of the inhomogeneous equation πx 1 , 0 < x < a, t > 0, utt − uxx = k sin 2 c a u(x, 0) = 0 = ut (x, 0) for 0 < x < a, u(0, t) = 0 = u(a, t) for t > 0. 34. Solve the Stokes problem which is concerned with the unsteady boundary layer flows induced in a semi-infinite viscous fluid bounded by an infinite horizontal disk at z = 0 due to nontorsional oscillations of the disk in its own plane with a given frequency ω. The equation of motion and the boundary and initial conditions are ut = νuzz , u(z, t) = U eiωt
z > 0, t > 0,
on z = 0, t > 0, u(z, t) → 0 as z → ∞ for t > 0, u(z, 0) = 0 for t ≤ 0 and z > 0, where u(z, t) is the velocity of the fluid of kinematic viscosity ν and U is constant. Solve the Rayleigh problem (ω = 0). Explain the physical significance of both the Stokes and Rayleigh solutions. 35. Solve the Blasius problem of an unsteady boundary layer flow in a semi-infinite body of viscous fluid enclosed by an infinite horizontal disk at z = 0. The governing equation and the boundary and initial conditions are ∂2u ∂u = ν 2 , z > 0, t > 0, ∂t ∂z u(z, t) = U t on z = 0, t > 0, u(z, t) → 0 as z → ∞, t > 0, u(z, t) = 0 at t ≤ 0, z > 0. Explain the significance of the solution. 36. Obtain the solution of the Stokes–Ekman problem of an unsteady boundary layer flow in a semi-infinite body of viscous fluid bounded by an infinite horizontal disk at z = 0 when both the fluid and the disk rotate with a uniform angular velocity Ω about the z-axis. The governing boundary layer equation and the boundary and the initial conditions are ∂2q ∂q + 2Ωiq = ν 2 , z > 0, t > 0, ∂t ∂z q(z, t) = aeiωt + be−iωt on z = 0, t > 0,
132
1 Linear Partial Differential Equations
q(z, t) → 0 as z → ∞ t > 0, q(z, t) = 0 at t ≤ 0, for all z > 0, where q = u + iv, ω is the frequency of oscillations of the disk, and a, b are complex constants. Hence, deduce the steady-state solution and determine the structure of the associated boundary layers. 37. Show that, when ω = 0 in Exercise 36, the steady-flow field is given by q(z, t) ∼ (a + b) exp
1/2 2iΩ z . − ν
Hence, determine the thickness of the Ekman layer. 38. Solve the telegraph equation utt − c2 uxx + 2aut = 0, u(x, 0) = 0,
−∞ < x < ∞, t > 0, ut (x, 0) = g(x),
−∞ < x < ∞.
39. Use the Laplace transform to solve the initial boundary-value problem ut = c2 uxx ,
0 < x < a, t > 0, 3πx u(x, 0) = x + sin for 0 < x < a, a u(0, t) = 0 = u(a, t) for t > 0.
40. Solve the diffusion equation −a < x < a, t > 0,
ut = κuxx ,
u(x, 0) = 1 for − a < x < a, u(−a, t) = 0 = u(a, t) for t > 0. 41. Use the joint Laplace and Fourier transform to solve the initial-value problem for transient water waves which satisfies (see Debnath 1994, p. 92) ∇2 φ = φxx + φzz = 0, −∞ < x < ∞, −∞ < z < 0, t > 0, φz = ηt , on z = 0, t > 0, φt + gη = − Pρ p(x)eiωt φ(x, z, 0) = 0 = η(x, 0) for all x and z, where P and ρ are constants. 42. Show that the solution of the boundary-value problem 1 urr + ur + uzz = 0, r u(r, z) = √
0 < r < ∞, 0 < z < ∞, 1 + r2
a2
on z = 0, 0 < r < ∞,
1.15 Exercises
is
∞
u(r, z) =
e−κ(z+a) J0 (κr) dκ =
0
133
1 . (z + a)2 + r2
43. (a) The axisymmetric initial-value problem is governed by 1 ut = κ urr + ur + δ(t)f (r), 0 < r < ∞, t > 0, r u(r, 0) = 0 for 0 < r < ∞. Show that the formal solution of this problem is ∞ kJ0 (kr)f˜(k) exp −k 2 κt dk. u(r, t) = 0
− r), show that the solution is Q ∞ J0 (kr)J1 (ak) exp −k 2 κt dk. u(r, t) = πa 0
(b) When f (r) =
Q πa2 H(a
44. If f (r) = A(a2 + r2 )− 2 , where A is a constant, show that the solution of the biharmonic equation described in Example 1.10.7 is 1
u(r, z) = A
{r2 + (z + a)(2z + a)} . [r2 + (z + a)2 ]3/2
45. Solve the axisymmetric biharmonic equation for the free vibration of an elastic disk 2 2 1 ∂ ∂ 2 b + u + utt = 0, 0 < r < ∞, t > 0, ∂r2 r ∂r u(r, 0) = f (r), ut (r, 0) = 0 for 0 < r < ∞, D where b2 = 2σh is the ratio of the flexural rigidity of the disk and its mass 2hσ per unit area. 46. Show that the zero-order Hankel transform solution of the axisymmetric problem 1 urr + ur + uzz = 0, 0 < r < ∞, −∞ < z < ∞, r lim r 2 u = 0, lim (2πr)ur = −f (z), −∞ < z < ∞, r→0
r→0
is 1 u ˜(k, z) = 4πk Hence, show that u(r, z) =
1 4π
∞ −∞
∞ −∞
exp −k|z − ζ| f (ζ) dζ.
r2 + (z − ζ)
− 12
f (ζ) dζ.
134
1 Linear Partial Differential Equations
47. Solve the nonhomogeneous diffusion problem 1 ut = κ urr + ur + Q(r, t), r u(r, 0) = f (r), 0 < r < ∞,
0 < r < ∞, t > 0,
where κ is a constant. 48. Solve the problem of the electrified unit disk in the (x, t)-plane with center at the origin. The electric potential u(r, z) is axisymmetric and satisfies the boundary-value problem 1 urr + ur + uzz = 0, 0 < r < ∞, 0 < z < ∞, r u(r, 0) = u0 , 0 ≤ r ≤ a, ∂u = 0, on z = 0 for a < r < ∞, ∂z u(r, z) → 0 as z → ∞ for all r, where u0 is constant. Show that the solution is 2u0 ∞ sin ak −kz e u(r, z) = J0 (kr) dk. π 0 k 49. Solve the axisymmetric surface wave problem in deep water due to an oscillatory surface pressure. The governing equations are 1 ∇2 φ = φrr + φr + φzz = 0, 0 ≤ r < ∞, −∞ < z ≤ 0, r φt + gη = − Pρ p(r) exp(iωt), on z = 0, t > 0, φz − ηt = 0 φ(r, z, 0) = 0 = η(r, 0),
for 0 ≤ r < ∞, and −∞ < z ≤ 0.
50. Solve the Neumann problem for the Laplace equation 1 urr + ur + uzz = 0, 0 < r < ∞, 0 < z ≤ ∞, r 1 uz (r, 0) = − 2 H(a − r), 0 < r < ∞, πa u(r, z) → 0 as z → ∞ for 0 < r < ∞. Show that lim u(r, z) =
a→0
− 1 1 2 r + z2 2 . 2π
51. Solve the Cauchy problem for the wave equation in a dissipating medium 1 2 utt + 2κut = c urr + ur , 0 < r < ∞, t > 0, r u(r, 0) = f (r), ut (r, 0) = g(r) for 0 < r < ∞, where κ is a constant.
1.15 Exercises
135
52. Use the joint Laplace and Hankel transform to solve the initial boundary-value problem 1 2 c urr + ur + uzz = utt , 0 < r < ∞, 0 < z < ∞, t > 0, r uz (r, 0, t) = H(a − r)H(t), 0 < r < ∞, t > 0, u(r, z, t) → 0 as r → ∞ and u(r, z, t) = 0 = ut (r, z, 0),
u(r, z, t) → 0
as z → ∞,
and show that
∞ z z2 J1 (ak)J0 ck t2 − 2 J0 (kr) dk. ut (r, z, t) = −acH t − c c 0
53. Find the steady temperature u(r, z) in a beam 0 ≤ r < ∞, 0 ≤ z ≤ a, when the face z = 0 is kept at temperature u(r, 0) = 0 and the face z = a is insulated except that heat is supplied through a circular hole such that uz (r, a) = H(b − r). The temperature u(r, z) satisfies the axisymmetric equation 1 urr + ur + uzz = 0, r
0 ≤ r < ∞, 0 ≤ z ≤ a.
54. Find the integral solution of the initial boundary-value problem 1 urr + ur + uzz r u(r, z, 0) ∂u ∂r r=0 ∂u ∂z z=0 u(r, z, t)
= ut ,
0 ≤ r < ∞, 0 ≤ z < ∞, t > 0,
= 0,
for all r and z,
= 0,
for 0 ≤ z < ∞, t > 0,
H(a − r) = −√ , for 0 < r < ∞, 0 < t < ∞, a2 − r 2 → 0 as r → ∞ or z → ∞.
55. Solve the Cauchy–Poisson wave problem (Debnath 1989) for a viscous liquid of finite or infinite depth governed by the equations, free surface, boundary and initial conditions 1 φrr + φr + φzz = 0, r 1 1 ψt = ν ψrr + ψr − 2 ψ + ψzz , r r where φ(r, z, t) and ψ(r, z, t) represent the potential and stream functions, respectively, 0 ≤ r < ∞, −h ≤ z ≤ 0 (or −∞ < z ≤ 0), and t > 0.
136
1 Linear Partial Differential Equations
The free surface conditions are
⎫ ηt − w = 0, ⎪ ⎬ μ(uz − wr ) = 0, ⎪ ⎭ φt + gη + 2νwz = 0
on z = 0, t > 0,
where η = η(r, t) is the free surface elevation, u = φr + ψz and w = φz − ψr − ψr are the radial and vertical velocity components of liquid particles, μ = ρν is the dynamic viscosity, ρ is the density, and ν is the kinematic viscosity of the liquid. The boundary conditions at the rigid bottom are u = φr + ψz = 0, on z = −h. w = φz − 1r (rψ)r = 0 The initial conditions are η=a
δ(r) , r
φ = ψ = 0 at t = 0,
where a is a constant and δ(r) is the Dirac delta function. If the liquid is of infinite depth, the bottom boundary conditions are (φ, ψ) → (0, 0)
as z → ∞.
56. Use the joint Hankel and Laplace transform method to solve the initial boundary-value problem 1 δ(r) δ(t), 0 < r < ∞, t > 0, urr + ur + utt − 2εut = a r r u(r, t) → 0 as r → ∞, u(0, t) is finite for t > 0, u(r, 0) = 0 = ut (r, 0) for 0 < r < ∞. 57. Surface waves are generated in an inviscid liquid of infinite depth due to an explosion (Sen 1963) above it which generates the pressure field p(r, t). The velocity potential φ(r, z, t) satisfies the Laplace equation 1 urr + ur + uzz = 0, r
0 < r < ∞, t > 0,
and the free surface condition 1 ∂p utt + guz = H(r) − H r, r0 (t) ρ ∂t
on z = 0,
where ρ is the constant density of the liquid, r0 (t) is the extent of the blast, and the liquid is initially at rest. Solve this problem.
1.15 Exercises
137
58. Use the joint Laplace and Fourier transform to show that the solution of the inhomogeneous diffusion problem ut − κuxx = q(x, t), x ∈ R, t > 0, u(x, 0) = f (x) for all x ∈ R, can be expressed in terms of Green’s function G(x, t; ξ, τ ) as ∞ t ∞ dτ q(ξ, τ )G(x, t; ξ, τ ) dξ + f (ξ)G(x, t; ξ, 0) dξ, u(x, t) = 0
−∞
−∞
where G(x, t; ξ, τ ) =
(x − ξ)2 exp − . 4κ(t − τ ) 4πκ(t − τ ) 1
59. Find Green’s function G(x, t) of the Bernoulli–Euler equation on an elastic foundation EI
∂2G ∂4G + κG + m = W δ(x)δ(t), ∂x4 ∂t2
x ∈ R, t > 0,
with initial conditions G(x, 0) = 0 = Gt (x, 0). 60. Solve the initial boundary-value problem ut − κuxx = q(t)δ(x − V t), x ∈ R, t > 0, u(x, 0) = 0 and u(x, t) → 0 as |x| → ∞, where q(t) = 0 for t < 0 and V is constant. 61. Find the Green function satisfying the equation Gxx + Gyy = δ(x − ξ)δ(y − η), 0 < x, ξ < a, 0 < y, η < b G(x, y) = 0 on x = 0, and x = a; G(x, y) = 0 on y = 0 and y = b. 62. Show that the solution of the two-dimensional diffusion equation ut − κ(uxx + uyy ) = q(x, y, t); −∞ < x, y < ∞, t > 0, with u(x, y; 0) = 0 is
(x − ξ)2 + (y − η)2 exp − 4κ(t − τ ) 0 −∞
q(ξ, η, τ ) × dξ dη . 4πκ(t − τ )
u(x, y, t) =
t
dτ
∞
138
1 Linear Partial Differential Equations
63. Find the Green function for the one-dimensional Klein–Gordon equation utt − c2 uxx + d2 u = p(x, t),
x ∈ R, t > 0,
with the initial and boundary conditions u(x, 0) = 0 = ut (x, 0)
for all x ∈ R,
where c and d are constants. Show that Green’s function for this problem reduces to that of the wave equation in the limit as d → 0. Derive the Green functions for both two- and three-dimensional Klein–Gordon equations. 64. Use the Fourier series method to solve the equation for a diffusion model ut = κuxx ,
− < x < , t > 0,
with the periodic boundary conditions u(−, t) = u(, t), ux (−, t) = ux (, t)
t > 0,
and the initial condition u(x, 0) = f (x),
− ≤ x ≤ .
65. (a) Verify that √ un (x, y) = exp ny − n sin nx, where n is a positive integer, is the solution of the Cauchy problem for the Laplace equation uxx + uyy = 0, u(x, 0) = 0,
x ∈ R, y > 0, √ uy (x, 0) = n exp − n sin nx.
(b) Show that this Cauchy problem is not well posed. 66. Find the eigenvalues and eigenfunctions of the Sturm–Liouville problems (a) u + λu = 0, 0 < x < 1, u(0) = 0, u(1) + u (1) = 0, (c) u + λu = 0, 0 < x < 1,
u(0) = u(1), u (0) = u (1),
(b) u + λu = 0, 0 < x < 1, u (0) + u (1) = 0, (d) u + 2u + 4λu = 0, 0 < x < a, u(0) = 0 = u(a).
67. Show that the equation a2 (x)u + a1 (x)u + a0 (x) + λ u = 0,
1.15 Exercises
139
can be reduced into the Sturm–Liouville form d p(x)u + q(x) + λρ(x) u = 0, dx where
a1 (x) dx , p(x) = exp a2 (x)
q(x) =
p(x)a0 (x) , a2 (x)
ρ(x) =
p(x) . a2 (x)
68. Reduce the given equation into the Sturm–Liouville form (a) u − 2xu + λxu = 0,
(b) u + u + λu = 0,
(c) xu + (1 − x)u + λu = 0 (x > 0), (d) x2 (x2 + 1)u + 2x3 u + λu = 0, x > 0. 69. Determine the Euler load and the corresponding fundamental buckling mode of a simply supported beam of length a under an axial compressive force P which is governed by the eigenvalue problem y (4) (x) + λy = 0, 0 < x < a, y(0) = 0 = y (0), y(a) = 0 = y (a), P ) > 0. where λ = ( EI 70. Use the Fourier method to solve the Klein–Gordon problem
utt − c2 uxx + d2 u = 0, 0 < x < a, t > 0, u(0, t) = 0 = u(a, t), t > 0, u(x, 0) = f (x),
ut (x, 0) = 0,
0 < x < a.
71. Use the Fourier method to solve the diffusion model ut = uxx + 2bux , 0 < x < a, t > 0, u(0, t) = 0 = u(a, t), t > 0, u(x, 0) = f (x), 0 < x < a. 72. (a) Solve the vibration problem of a circular membrane governed by (1.13.56) and (1.13.59), (1.13.60) with the boundary and initial conditions (1.13.52), (1.13.53) when g(r, θ) = 0. (b) Obtain the solution for the problem (a) when f (r, θ) = a2 − r 2 . 73. (a) Use the method of separation of variables to solve the Dirichlet problem in the cylinder for u(r, z) 1 urr + ur + uzz = 0, 0 ≤ x ≤ a, 0 ≤ z ≤ h, r u(r, 0) = 0 = u(a, z), u(r, h) = f (r). (b) If f (r) = 1, obtain the solution.
140
1 Linear Partial Differential Equations
74. (a) Derive the differential equality 2ut c2 ∇2 u − utt = 2c2 (ut ux )x + (ut uy )y − c2 u2x + u2y + u2t t , associated with the wave equation c2 (uxx + uyy ) − utt ≡ c2 ∇2 u − utt = 0. (b) Generalize the above differential equality for the (n + 1)-dimensional wave equation (1.13.9). (c) Show that the differential equality for the (n + 1)-dimensional wave equation can be written in the form 2ut c2 ∇2n − utt = 2c2 ∇n · (ut ∇n u) − c2 |∇n u|2 + u2t t , where
∇n =
∂ ∂ ∂ . , ,..., ∂x1 ∂x2 ∂xn
75. (a) Derive the energy integral E(t) =
1 2
b
u2t + c2 u2x + d2 u2 dx,
a
for the Klein–Gordon equation model utt − c2 uxx + d2 u = 0, a ≤ x ≤ b, t > 0, u(a, t) = 0 = u(b, t), u(x, 0) = f (x), ut (x, 0) = g(x),
a ≤ x ≤ b.
(b) Show that the energy is constant, that is, E(t) = E0 = const. (c) Use the law of conservation of energy to prove that the Klein–Gordon equation has a unique solution. 76. (a) Use the method of separation of variables to solve the spherically symmetric wave equation (1.13.21) in three dimensions 2 utt = c2 ∇2 u ≡ c2 urr + ur , 0 < r < a, t > 0, r u(a, t) = 0, t > 0, u(r, 0) = f (r), ut (r, 0) = g(r), 0 < r < a. (b) Find the solution for f (r) = 0 and g(r) = a − r.
1.15 Exercises
141
77. Solve the axisymmetric Dirichlet problem in a right circular cylinder urr + r−1 ur + uzz ≡ 0,
0 < r < a, 0 < z < h,
u(r, 0) = 0 = u(r, h),
u(a, z) = f (z).
78. Use Example 1.6.4 to find the solution of equation (1.6.80) with the boundary conditions (a) u(1, θ) = 2 cos2 θ,
(b) u(1, θ) = |2θ|,
(c) ur (1, θ) = 2 cos 2θ,
(d) ur (1, θ) = cos θ + sin θ.
79. Apply the method of separation of variables to solve the Laplace equation in an annular region 1 1 urr + ur + 2 uθθ = 0, r r
0 < a < r < b, 0 ≤ θ ≤ 2π
with the following boundary conditions: (a) u(a, θ) = f (θ), u(b, θ) = g(θ), 0 ≤ θ ≤ 2π. (b) u(1, θ) = 12 + sin θ, u(2, θ) = 12 (1 − 1n2 + 2 cos θ), 0 ≤ θ ≤ 2π. (c) ur (1, θ) = 0, ur (2, θ) = 34 (cos θ − sin θ), 0 ≤ θ ≤ 2π. 80. In Example 1.6.1, take c = 1, = π. If u(x, t) = 12 [f (x − t) + f (x + t)] + x+t g(τ ) dτ , then show that x−t
π
u2x + u2t dx =
0
π
f 2 + g 2 dx,
0
where f and g are real functions on 0 ≤ x ≤ π with continuous partial derivatives and f (0) = f (π) = g(0) = g(π) = 0. 81. Consider the boundary-value problem for the elliptic equation |x| |y| 2 < 1 , + ∇ u + 1 = 0 in D = (x, y) a b u = 0 on ∂D, where a > b > 0. Show that a2 a 2 b2 ≤ u(0, 0) ≤ . 4(a2 + b2 ) 4 82. Show that the Dirichlet problem ∇2 u = 0, x ∈ D, u(x) = f (x) on ∂D, has a unique solution.
142
1 Linear Partial Differential Equations
83. Consider the Dirichlet boundary-value problem |x| |y| 2 + <1 , ∇ u + 1 = 0, x ∈ (x, y) a b u = 0 on ∂D, where a > b > 0. Use a suitable function v(x, y) = Ax2 + By 2 satisfying ∇2 v = 1 with (A, B) > (0, 0) to prove that (ab)2 (ab)2 . ≤ u(0, 0) ≤ 2 2(a + b) 2(a2 + b2 ) 84. Solve the fractional Blasius problem as stated in Exercise 35 with the governing equation (see Debnath 2003a, 2003b) ∂ 2u ∂αu = ν 2, α ∂t ∂z
x ∈ R, t > 0.
85. Solve the fractional Stokes–Ekman problem as stated in Exercise 36 with the governing equation (see Debnath 2003a, 2003b) ∂αq ∂ 2q + 2iΩq = ν 2 , α ∂t ∂z
x ∈ R, t > 0.
86. Apply the method of separation of variables u(x, t) = X(x)T (t) to solve the eigenvalue problem for the dissipation wave equation utt − c2 uxx + αut = 0, 0 < x < , t > 0, u(0, t) = 0 = u(, t), t > 0, u(x, 0) = f (x), ut (x, 0) = g(x), 0 < x < . Show that (i) X + λ2 X = 0, T¨ + αT˙ + λ2 c2 T = 0, where λ2 is a separation constant. nπx (ii) The eigenvalues and eigenfunctions are λn = nπ , Xn (x) = Bn sin( ), n = 1, 2, . . . . (iii) The general solution is u(x, t) =
1 1 1 e− 2 αt an cos δn t + bn sin δn t , 2 2 n=1 ∞
where δn = 4λ2n c2 − α2 . Consider other cases 4λ2n c2 < or = α2 . 87. Solve the above problem 86 with the same initial conditions and the following boundary conditions: (i) ux (0, t) = 0 = u(, t), t > 0, (ii) u(0, t) = 0 = ux (, t), t > 0, Find the eigenvalues, eigenfunctions, and the general solution in each case.
1.15 Exercises
88. Solve the eigenvalue problem for the telegraph equation utt − c2 uxx + aut + b = 0,
0 < x < 1, t > 0,
with the boundary and initial conditions u(0, t) = 0, ux (1, t) + u(1, t) = 0, t > 0, u(x, 0) = f (x) and ut (x, 0) = g(x), 0 < x < 1. Show that the eigenvalue equation and the eigenfunctions are and
tan λ = λ
Xn (x) = An sin λx,
where −λ2 is the separation constant. 89. Use the method of separation of variables to solve the problem utt − c2 uxx + αu = 0, 0 < x < 1, t > 0, ux (0, t) = 0 = u(1, t), t > 0, u(x, 0) = f (x), ut (x, 0) = g(x),
0 < x < 1.
Show that the eigenvalues and the eigenfunctions are
12 πx 1 2 2 2 , (2n − 1) π c + α , λn = Xn (x) = An cos (2n − 1) 4 2 where n = 1, 2, 3, . . . , and −λ2 is the separation constant. Derive the general solution
∞ πx u(x, t) = . (an cos λn t + bn sin λn t) cos (2n − 1) 2 n=1 Show that the solution for f (x) = x and g(x) = 0 corresponds to
4 2 n−1 (−1) , bn = 0. − an = (2n − 1)π (2n − 1)π 90. Consider the eigenvalue problem utt − c2 uxx + au = 0,
0 < x < 1, t > 0,
ux (0, t) = 0, ux (1, t) + u(1, t) = 0, t > 0, with the initial conditions u(x, 0) = f (x)
and
ut (x, 0) = g(x),
0 < x < 1,
(a) Show that, for u(x, t) = X(x)T (t), c2 X − aX + λ2 X = 0,
T¨ + λ2 T = 0;
and the eigenvalue equation is β tan β = 1,
β=
where −λ2 is the separation constant.
1 1 2 λ − a 2, c
143
144
1 Linear Partial Differential Equations
91. Consider the Helmholtz equation 1 1 urr + ur + 2 uθθ + ε2 u = 0 for r ≤ 1, r r where ε =
ω c
1, with the boundary condition u(1, θ) = sin θ.
Obtain the asymptotic solution in the form u = u0 + ε2 u2 + O ε4 , where u = O(1) on the boundary. Show that the two-term asymptotic solution is 1 u(r, θ) = r sin θ + ε2 r − r3 + O ε4 . 8 92. Consider the boundary value problem for the modified Helmholtz equation ε2 ∇2 u = u, with u(1, θ) = 1 and u(r, θ) → 0 as r → ∞. Show that the asymptotic solution is given by 1−r . u = exp ε 93. (a) The temperature distribution u(x, t) in a homogeneous rod of length with insulated endpoints is described by the initial boundary-value problem ut = κuxx , 0 < x < , t > 0, ux (0, t) = 0 = ux (, t), t > 0, u(x, 0) = f (x), 0 < x < . (b) Obtain the solution from (a) for the case = 1, f (x) = x. 94. (a) Show that the telegraph equation can be written in the form utt − c2 uxx + (p + q)ut + pqu = 0, where p = (G/C) and q = (R/L). (b) Apply the transformation u = v exp[− 12 (p + q)t] to transform the equation in the form vtt − c2 vxx =
1 (p − q)2 v. 4
(c) When p = q, there exists an undistorted wave solution. Show that a progressive wave of the form
1.15 Exercises
145
u(x, t) = exp(−pt)f (x ± ct) propagate in either direction, where f is an arbitrary twice differentiable function of its argument. (d) If u(x, t) = A exp[i(kx − ωt)] is a solution of equation as stated in (a), show that the dispersion relation is ω 2 + i(p + q)ω − c2 k 2 + pq = 0. 95. Consider the telegraph equation problem ut − c2 uxx + aut + bu = 0, 0 < x < l, t > 0, u(x, 0) = f (x), ut (x, 0) = g(x) for 0 < x < l, u(0, t) = 0 = u(l, t)
for t ≥ 0,
where a and b are constants. (a) Show that, for any T > 0 l t 2 2 2 2 2 ut + c ux + bu t=T dx ≤ ut + c2 u2x + bu2 t=0 dx. 0
0
(b) Use the above integral inequality from (a) to show that the initial boundaryvalue problem for the telegraph equation can have only one solution. 96. Use the solution (1.9.15) to obtain the solution of the nonhomogeneous wave equation problem utt − c2 uxx = sin(kx − ωt), x ∈ R, t > 0, u(x, 0) = 0 = ut (x, 0), x ∈ R. Discuss the solution for cases (a) c = ωk and (b) c = ωk (resonance). 97. Derive the Duhamel formula for the solution of Example 1.9.2 is t ∂u0 ∂ f (t − τ ) dτ u(x, t) = f (t) ∗ u0 (x, t) = ∂t ∂τ 0 where u0 (x, t) = L−1
x s 1 = erfc √ exp −x . s k 4κt
98. Solve the axisymmetric unsteady viscous flow problem in a long rotating cylinder of radius a governed by 1 v vt = ν vrr + vr − 2 , 0 < r ≤ a, t > 0, r r where v = v(r, t) is the tangential fluid velocity and ν is the kinematic viscosity of the fluid.
146
1 Linear Partial Differential Equations
The cylinder is at rest until at t = 0+, it is caused to rotate so that the boundary and initial conditions are v(r, t) = aΩf (t)H(t) on r = a, v(r, 0) = 0 for r < a, where f (t) is a physically realistic function of time t. Find the solution when f (t) = cos ωt and examine its feature as t → ∞. 99. (a) Show that the solution of the Cauchy problem for the diffusion equation ut = uxx , x ∈ R, t > 0, 2 u(0, t) = sin 2n2 t , ux (0, t) = 0, n
t > 0,
is given by un (x, t) =
1 nx e sin nx + 2n2 t − e−nx sin nx − 2n2 t n
where n is an integer. (b) Show that this Cauchy problem is ill-posed. 100. (a) Verify that √ un (x, y) = exp ny − n sin nx is the solution of the Cauchy problem for the Laplace equation uxx + uyy = 0, u(x, 0) = 0,
x ∈ R, y > 0, √ uy (x, 0) = n exp − n sin nx,
where n is an integer. (b) Show that this Cauchy problem is ill-posed. 101. (a) Verify that un (x, y) =
1 −√ n e sin nx sinh ny n
is the solution of the Cauchy problem for the Laplace equation in the upper half-strip uxx + uyy = 0, 0 ≤ x ≤ π, y > 0, u(0, y) = 0 = u(π, y), y > 0 u(x, 0) = 0 and
uy (x, 0) = e−
√ n
sin nx.
(b) Show that this Cauchy problem is ill-posed. 102. Using the wave function ψ(x, t) = a(x, t) exp[ hi S(x, t)], where a and S are real functions and the transformation u(x, t) = m−1 Sx in the Schrödinger equation (1.7.66), derive
1.15 Exercises
147
∂u ∂u ∂ 2 1 +u = V (x)a + axx , ∂t ∂x ∂x 2m a 2 ∂a ∂ + = 0. ∂t ∂x (Hydrodynamic analogy of quantum mechanics). 103. Consider the two-dimensional boundary value problem for the Laplace equation in the upper half plane uxx + uyy = 0, x ∈ R, y > 0, u(x, 0) = f (x) and uy (x, 0) = g(x),
x ∈ R.
(a) If f (x) = 0 and g(x) = 0, show that u(x, y) ≡ 0 is the solution. (b) If the boundary data is changed to u(x, 0) = n1 cos nx and g(x) = 0, x ∈ R, show that u(x, y) = n1 cos nx cosh ny is the solution. Examine the ill-posedness of the problem. 104. (a) Show that the solution of the Cauchy problem for the negative diffusion equation ut + κuxx = 0,
x ∈ R, t > 0, κ > 0,
u(x, 0) = f (x),
x∈R
is given by u(x, t) =
1 exp κn2 t sin nx. n
(b) Show that this negative diffusion problem is ill-posed. 105. (a) Show that the ill posed problem 104(a) can be made well posed by adding higher order diffusive terms, that is, the modified Cauchy problem ut + κuxx + δuxxxx = 0, x ∈ R, t > 0 (κ, δ > 0), 1 u(x, 0) = sin nx, x ∈ R n is a well posed problem. (b) Verify that u(x, t) =
1 sin nx exp κn2 − δn4 t n
is the solution of the modified Cauchy problem.
2 Nonlinear Model Equations and Variational Principles
True Laws of Nature cannot be linear. Albert Einstein . . . the progress of physics will to a large extent depend on the progress of nonlinear mathematics, of methods to solve nonlinear equations . . . and therefore we can learn by comparing different nonlinear problems. Werner Heisenberg Our present analytical methods seem unsuitable for the solution of the important problems arising in connection with nonlinear partial differential equations and, in fact, with virtually all types of nonlinear problems in pure mathematics. The truth of this statement is particularly striking in the field of fluid dynamics. . . John Von Neumann
2.1 Introduction This chapter deals with the basic ideas and many major nonlinear model equations which arise in a wide variety of physical problems. Included are one-dimensional wave, Klein–Gordon (KG), sine–Gordon (SG), Burgers, Fisher, Korteweg–de Vries (KdV), Boussinesq, modified KdV, nonlinear Schrödinger (NLS), Benjamin–Ono (BO), Benjamin–Bona–Mahony (BBM), Ginzburg–Landau (GL), Burgers–Huxley (BH), KP, concentric KdV, Whitham, Davey–Stewartson, Toda lattice, Camassa– Holm (CH), and Degasperis–Procesi (DP) equations. This is followed by variational principles and the Euler–Lagrange equations. Also included are Plateau’s problem, Hamilton’s principle, Lagrange’s equations, Hamilton’s equations, the variational principle for nonlinear Klein–Gordon equations, and the variational principle for nonlinear water waves. Special attention is given to the Euler equation of motion, L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_2,
150
2 Nonlinear Model Equations and Variational Principles
the continuity equation, the associated energy equation and energy flux, linear water wave problems and their solutions, nonlinear finite amplitude waves (the Stokes waves), gravity waves, gravity-capillary waves, and linear and nonlinear dispersion relations. Finally, the modern theory of nonlinear water waves is formulated.
2.2 Basic Concepts and Definitions The most general first-order nonlinear partial differential equation in two independent variables x and y has the form F (x, y, u, ux , uy ) = 0.
(2.2.1)
The most general second-order nonlinear partial differential equation in two independent variables x and y has the form F (x, y, u, ux , uy , uxx , uxy , uyy ) = 0.
(2.2.2)
Similarly, the most general first-order and second-order nonlinear equations in more independent variables can be introduced. More formally, it is possible to write these equations in the operator form Lx u(x) = f (x),
(2.2.3)
where Lx is a partial differential operator and f (x) is a given function of two or more independent variables x = (x, y, . . .). It has already been indicated in Section 1.2 that if Lx is not a linear operator, (2.2.3) is called a nonlinear partial differential equation. Equation (2.2.3) is called an inhomogeneous nonlinear equation if f (x) = 0. On the other hand, (2.2.3) is called a homogeneous nonlinear equation if f (x) = 0. In general, the linear superposition principle can be applied to linear partial differential equations if certain convergence requirements are satisfied. This principle is usually used to find a new solution as a linear combination of a given set of solutions. For nonlinear partial differential equations, however, the linear superposition principle cannot be applied to generate a new solution. So, because most solution methods for linear equations cannot be applied to nonlinear equations, there is no general method of finding analytical solutions of nonlinear partial differential equations, and numerical techniques are usually required for their solution. A transformation of variables can sometimes be found that transforms a nonlinear equation into a linear equation, or some other ad hoc method can be used to find a solution of a particular nonlinear equation. In fact, new methods are usually required for finding solutions of nonlinear equations. Methods of solution for nonlinear equations represent only one aspect of the theory of nonlinear partial differential equations. Like linear equations, questions of existence, uniqueness, and stability of solutions of nonlinear partial differential equations are of fundamental importance. These and other aspects of nonlinear equations have led the subject into one of the most diverse and active areas of modern mathematics.
2.3 Some Nonlinear Model Equations
151
2.3 Some Nonlinear Model Equations Nonlinear partial differential equations arise frequently in formulating fundamental laws of nature and in the mathematical analysis of a wide variety of physical problems. Listed below are some important model equations of most common interest. Example 2.3.1. The simplest first-order nonlinear wave (or kinematic wave) equation is (2.3.1) ut + c(u)ux = 0, x ∈ R, t > 0, where c(u) is a given function of u. This equation describes the propagation of a nonlinear wave (or disturbance). A large number of nonlinear problems governed by equation (2.3.1) include waves in traffic flow on highways (Lighthill and Whitham 1955; Richards 1956), shock waves, flood waves, waves in glaciers (Nye 1960, 1963), chemical exchange processes in chromatography, sediment transport in rivers (Kynch 1952), and waves in plasmas. Example 2.3.2. The nonlinear Klein–Gordon equation is utt − c2 ∇2 u + V (u) = 0,
(2.3.2)
where c is a constant, and V (u) is a nonlinear function of u usually chosen as the derivative of the potential energy V (u). It arises in many physical problems including nonlinear dispersion (Scott 1969; Whitham 1974) and nonlinear meson theory (Schiff 1951). Example 2.3.3. The sine-Gordon equation utt − c2 uxx + κ sin u = 0,
x ∈ R, t > 0,
(2.3.3)
where c and κ are constants, has arisen classically in the study of differential geometry, and in the propagation of a ‘slip’ dislocation in crystals (Frenkel and Kontorova 1939). More recently, it arises in a wide variety of physical problems including the propagation of magnetic flux in Josephson-type super conducting tunnel junctions, the phase jump of the wave function of superconducting electrons along long Josephson junctions (Josephson 1965; Scott 1969), a chain of rigid pendula connected by springs (Scott 1969), propagation of short optical pulses in resonant laser media (Arecchi et al. 1969; Lamb 1971), stability of fluid motions (Scott et al. 1973; Gibbon 1985), in ferromagnetism and ferroelectric materials, in the dynamics of certain molecular chains such as DNA (Barone et al. 1971), in elementary particle physics (Skyrme 1958, 1961; Enz 1963), and in weakly unstable baroclinic wave packets in a two-layer fluid (Gibbon et al. 1979). Example 2.3.4. The Burgers equation is ut + uux = νuxx ,
x ∈ R, t > 0,
(2.3.4)
where ν is the kinematic viscosity. This is the simplest nonlinear model equation for diffusive waves in fluid dynamics. It was first introduced by Burgers (1948) to
152
2 Nonlinear Model Equations and Variational Principles
describe one-dimensional turbulence, and it also arises in many physical problems including sound waves in a viscous medium (Lighthill 1956), waves in fluid-filled viscous elastic tubes, and magnetohydrodynamic waves in a medium with finite electrical conductivity. Example 2.3.5. The Fisher equation u2 , ut − νuxx = k u − κ
x ∈ R, t > 0,
(2.3.5)
where ν, k, and κ are constants, is used as a nonlinear model equation to study wave propagation in a large number of biological and chemical systems. Fisher (1936) first introduced this equation to investigate wave propagation of a gene in a population. It is also used to study logistic growth–diffusion phenomena. In recent years, the Fisher equation has been used as a model equation for a large variety of problems which include gene-culture waves of advance (Aoki 1987), chemical wave propagation (Arnold et al. 1987), neutron population in a nuclear reactor (Canosa 1969, 1973), and spread of early farming in Europe (Ammerman and Cavalli-Sforva 1971). It also arises in the theory of combustion, nonlinear diffusion, and chemical kinetics (Kolmogorov et al. 1937; Aris 1975; and Fife 1979). Example 2.3.6. The Boussinesq equation utt − uxx + 3u2 xx − uxxxx = 0
(2.3.6)
describes one-dimensional weakly nonlinear dispersive water waves propagating in both positive and negative x-directions (Peregrine 1967; Toda and Wadati 1973; Zakharov 1968a, 1968b; Ablowitz and Haberman 1975; and Prasad and Ravindran 1977). It also arises in one-dimensional lattice waves (Zabusky 1967) and ionacoustic solitons (Kako and Yajima 1980). In recent years, considerable attention has been given to new forms of Boussinesq equations (Madsen et al. 1991; Madsen and Sorensen 1992, 1993) dealing with water wave propagation and to modified Boussinesq equations (Nwogu 1993; Chen and Liu 1995a, 1995b) in terms of a velocity potential on an arbitrary elevation and free surface displacement of water. Example 2.3.7. The Korteweg–de Vries (KdV) equation ut + αuux + βuxxx = 0,
x ∈ R, t > 0,
(2.3.7)
where α and β are constants, is a simple and useful model for describing the long time evolution of dispersive wave phenomena in which the steepening effect of the nonlinear term is counterbalanced by the dispersion. It was originally introduced by Korteweg and de Vries (1895) to describe the propagation of unidirectional shallow water waves. It admits the exact solution called the soliton. This equation arises in many physical problems including water waves (Johnson 1980, 1997; Debnath 1994), internal gravity waves in a stratified fluid (Benney 1966; Redekopp and Weidman 1968),
2.3 Some Nonlinear Model Equations
153
ion-acoustic waves in a plasma (Washimi and Taniuti 1966), pressure waves in a liquid-gas bubble (Van Wijngaarden 1968), and rotating flow in a tube (Leibovich 1970). There are other physical systems to which the KdV equation applies as a long wave approximation, including acoustic-gravity waves in a compressible heavy liquid, axisymmetric waves in a nonuniformly rotating fluid, acoustic waves in anharmonic crystals, nonlinear waves in cold plasmas, axisymmetric magnetohydrodynamic waves, and longitudinal dispersive waves in elastic rods. Example 2.3.8. The modified KdV (mKdV) equation ut − 6u2 ux + uxxx = 0,
x ∈ R, t > 0,
(2.3.8)
describes nonlinear acoustic waves in an anharmonic lattice (Zabusky 1967) and Alfvén waves in a collisionless plasma (Kakutani and Ono 1969). It also arises in many other physical situations. Example 2.3.9. The nonlinear Schrödinger (NLS) equation iut + uxx + γ|u|2 u = 0,
x ∈ R, t > 0,
(2.3.9)
where γ is a constant, describes the evolution of water waves ((Benney and Roskes 1969; Hasimoto and Ono 1972; Davey 1972; Davey and Stewartson 1974; Peregrine 1983); Zakharov 1968a, 1968b; Chu and Mei 1970; Yuen and Lake 1975; Infeld et al. 1987; Johnson 1997). It also arises in some other physical systems which include nonlinear optics (Kelley 1965; Talanov 1965; Bespalov and Talanov 1966; Karpman and Krushkal 1969; Asano et al. 1969; Hasegawa and Tappert 1973), hydromagnetic and plasma waves (Ichikawa et al. 1972; Schimizu and Ichikawa 1972; Taniuti and Washimi 1968; Fulton 1972; Hasegawa 1990; Ichikawa 1979; Weiland and Wiljelmsson 1977; Weiland et al. 1978), the propagation of a heat pulse in a solid (Tappert and Varma 1970), nonlinear waves in a fluid-filled viscoelastic tube (Ravindran and Prasad 1979), nonlinear instability problems (Stewartson and Stuart 1971; Nayfeh and Saric 1971), and the propagation of solitary waves in piezoelectric semiconductors (Pawlik and Rowlands 1975). Example 2.3.10. The Benjamin–Ono (BO) equation is ut + uux + H{uxx } = 0, where H{f (ξ, t)} = f˜(x, t) is the Hilbert transform of f (ξ, t) defined by ∞ 1 f (ξ, t) dξ , H f (ξ, t) = P π ξ−x −∞
(2.3.10)
(2.3.11)
where P stands for the Cauchy principal value. This equation arises in the study of weakly nonlinear long internal gravity waves (Benjamin 1967; Davis and Acrivos 1967; and Ono 1975) and belongs to the class of weakly nonlinear models.
154
2 Nonlinear Model Equations and Variational Principles
Example 2.3.11. The Benjamin–Bona–Mahony (BBM) equation ut + ux + uux − uxxt = 0,
x ∈ R, t > 0,
(2.3.12)
represents another nonlinear model for long water waves. The KdV equation can be written as ut + ux + uux + uxxx = 0, x ∈ R, t > 0. (2.3.13) The basic mathematical difference between the BBM and KdV equations can readily be determined by comparing the approximate dispersion relations for the respective linearized equations. We seek a plane wave solution of both linearized equations of the form
u(x, t) ∼ exp i(ωt − kx) . (2.3.14) The dispersion relation of the linearized KdV equation is then given by ω = k − k3 .
(2.3.15)
The phase and group velocities are given by Cp =
ω = 1 − k2 k
and
Cg =
dω = 1 − 3k 2 , dk
(2.3.16ab)
which become negative for k 2 > 1. This means that all waves of large wavenumbers (small wavelengths) propagate in the negative x-direction in contradiction to the original assumption that waves travel only in the positive x-direction. This is an undesirable physical feature of the KdV equation. To eliminate this unrealistic feature of the KdV equation, Benjamin et al. (1972) proposed equation (2.3.12). The dispersion relation of the linearized version of (2.3.12) is ω=
k . (1 + k2 )
(2.3.17)
Thus the phase and group velocities of waves associated with this model are given by −1 −2 ω Cp = = 1 + k 2 , Cg = 1 − k 2 1 + k 2 . (2.3.18ab) k Both Cp and Cg tend to zero, as k → ∞, showing that short waves do not propagate. In other words, the BBM model has the approximate features of responding only significantly to short wave components introduced in the initial wave form. Thus, the BBM equation seems to be a preferable model. However, the fact that the BBM model is a better model than the KdV model has not been fully confirmed, yet. Example 2.3.12. The Ginzburg–Landau (GL) equation is At + aAxx = bA + cA|A|2 ,
(2.3.19)
where a and b are complex constants determined by the dispersion relation of linear waves, and c is determined by the weakly nonlinear interaction (Stewartson and Stuart 1971). This equation describes slightly unstable nonlinear waves and has arisen originally in the theories of superconductivity and phase transitions.
2.3 Some Nonlinear Model Equations
155
The complex Ginzburg–Landau equation simplifies significantly if all of the coefficients are real. The real Ginzburg–Landau equation has been extensively investigated in problems dealing with phase separation in condensed matter physics (BenJacob et al. 1985; Van Saarloos 1989; Balmforth 1995). Example 2.3.13. The Burgers–Huxley (BH) equation ut + αuux − νuxx = β(1 − u)(u − γ)u,
x ∈ R, t > 0,
(2.3.20)
where α, β ≥ 0, γ (0 < γ < 1), and ν are parameters, describes the interaction between convection, diffusion, and reaction. When α = 0, equation (2.3.20) reduces to the Hodgkin and Huxley (1952) equation which describes nerve pulse propagation in nerve fibers and wall motion in liquid crystals (Scott 1977; Satsuma 1987a, 1987b; Wang 1985, 1986; Wang et al. 1990). Because of the complexity of the Huxley equation, the FitzHugh–Nagumo equations (FitzHugh 1961; Sleeman 1982; Nagumo et al. 1962) proposed simple, analytically tractable, and particularly useful model equations which contain the key features of the Huxley model. On the other hand, when β = 0, equation (2.3.20) reduces to the Burgers equation (2.3.4) describing diffusive waves in nonlinear dissipating systems. Satsuma (1987a, 1987b) obtained solitary wave solutions of (2.3.20) by using Hirota’s method in soliton theory. Example 2.3.14. The Kadomtsev–Petviashvili (KP) equation (ut − 6uux + uxxx )x + 3σ 2 uyy = 0,
(2.3.21)
is a two-dimensional generalization of the KdV equation. Kadomtsev and Petviashivili (1970) first introduced this equation to describe slowly varying nonlinear waves in a dispersive medium (Johnson 1980, 1997). Equation (2.3.21) with σ 2 = +1 arises in the study of weakly nonlinear dispersive waves in plasmas and also in the modulation of weakly nonlinear long water waves (Ablowitz and Segur 1979) which travel nearly in one dimension (that is, nearly in a vertical plane). Satsuma (1987a, 1987b) showed that the KP equation has N line-soliton solutions which describe the oblique interaction of solitons. The equation with σ 2 = −1 arises in acoustics and admits unstable soliton solutions, whereas for σ 2 = +1 the solitons are stable. Freeman (1980) presented an interesting review of soliton interactions in two dimensions. Recently, Chen and Liu (1995a, 1995b) have derived the unified KP (uKP) equation for surface and interfacial waves propagating in a rotating channel with varying topography and sidewalls. This new equation includes most of the existing KP-type equations in the literature as special cases. Example 2.3.15. The concentric KdV equation 2uR +
1 1 u + 3uuξ + uξξξ = 0 R 3
(2.3.22)
describes concentric waves on the free surface of water that have decreasing amplitude with increasing radius. This is also called the cylindrical KdV equation which
156
2 Nonlinear Model Equations and Variational Principles
was first derived in another context by Maxon and Viecelli (1974). The inverse scattering transform for equation (2.3.22) involves a linearly increasing potential which yields eigenfunctions based on the Airy function (see Calogero and Degasperis 1978). A discussion of this equation and its solution can also be found in Johnson (1997) and Freeman (1980). Example 2.3.16. The nearly concentric KdV equation (or the Johnson equation) 1 1 1 (2.3.23) 2uR + u + 3uuξ + uξξξ + 2 uθθ = 0 R 3 R ξ describes the nearly concentric surface waves incorporating weak dependence on the angular coordinate θ. In the absence of θ-dependence, equation (2.3.23) reduces to (2.3.22). This equation was first derived by Johnson (1980) in his study of problems of nonlinear water waves. Example 2.3.17. The Davey–Stewartson (DS) equations −2ikcp Aτ + aAζζ − cp cg Ayy + bA|A|2 + ck2 Afζ = 0, 1 − c2g fζζ + fyy = d |A|2 ζ ,
(2.3.24) (2.3.25)
where a, b, c, d are functions of δk (see Davey and Stewartson 1974; Johnson 1997), describe weakly nonlinear dispersive waves propagating in the x-direction with a slowly varying structure in both the x- and y-directions. In the absence of y-dependence with fζ ≡ 0, the DS equations recover the NLS equation for water waves (see Hasimoto and Ono 1972) in the form −2ikcp Aτ + aAζζ + bA|A|2 = 0.
(2.3.26)
This is similar to (2.3.9). Example 2.3.18. The Whitham (1974) nonlinear nonlocal integrodifferential equation ∞ K(x − ξ)ηξ (ξ, t) dξ = 0 (2.3.27) ηt + dηηx + −∞
can describe symmetric waves that propagate without change of shape and peak at a critical height, as well as asymmetric waves that invariably break. The kernel K(x) is given by the inverse Fourier transform of the phase velocity c(k) = ωk in the form ∞ 1 K(x) = F −1 c(k) = eikx c(k) dk. (2.3.28) 2π −∞ It is a well known fact that the nonlinear shallow water equations which neglect dispersion altogether lead to breaking of the typical hyperbolic kind, with development of a vertical slope and a multivalued wave profile. It is clear that the third derivative dispersion term in the KdV equation (2.3.7) prevents wave breaking. Whitham formulated his equation (2.3.27) to describe the observed phenomena of solitary and periodic cnoidal waves as well as peaking and breaking of water waves. The Whitham
2.3 Some Nonlinear Model Equations
157
equation is a kind of generalization of the KdV equation that takes c(k) = c0 − γk 2 and K(x) = c0 δ(x) + δ (x), c20 = gh0 . The detailed analysis of Whitham’s analysis is given in Section 7.8 in Chapter 7. Example 2.3.19. The Camassa and Holm (CH) Equation for the free surface elevation u(x, t) over a flat rigid bottom is ut − uxxt + 3uux = 2ux uxx + uuxx ,
x ∈ R, t > 0.
(2.3.29)
It describes the propagation of nonlinear dispersive shallow water equation to capture the essential features of wave breaking. It is integrable in the sense that there exists a Lax pair, and has infinitely many conservation laws. The CH equation admits stable solitary wave solutions with a peak at their crests; these waves are called peakons. A more elaborate discussion of this equation (Camassa and Holm 1993) and its various extensions are presented in Section 9.13. Example 2.3.20. The Degasperis and Procesi (DP) equation is ut − uxxt + 4uux = 3ux uxx + uuxxx ,
x ∈ R, t > 0.
(2.3.30)
It also describes the propagation of nonlinear dispersive shallow water waves. Its solutions are singular, leading to wave breaking. The DP equation admits a shockpeakon solution which is significantly different from the peakon solutions of the CH equation. Both the CH and DP equations have soliton solutions which develop singularities in finite time (or solutions blow-up in finite time). Both the CH and DP equations can be combined into a (1 + 1)-dimensional b-family equation for fluid velocity u(x, t) in the form mt + umx + bmux = 0,
x ∈ R, t > 0,
(2.3.31)
where m = (u − uxx ) and u = g ∗ m is the convolution product given by 1 (2.3.32) u(x) = g(x − ξ)m(ξ) dξ, g(x) = exp −|x| , 2 R which determines the traveling wave shape and length scale for equation (2.3.31) and the constant b is a balance (or bifurcation) parameter. Degasperis and Procesi (1999) showed that (2.3.31) cannot be completely integrable unless b = 2 or b = 3. When b = 2, equation (2.3.31) reduces to the CH equation (2.3.29) and when b = 3, (2.3.31) becomes the DP equation (2.3.30). A more detailed discussion on these equations can be found in Section 9.13. Example 2.3.21 (The Toda Lattice Equation in a mass–spring system). A mass– spring lattice is an infinite chain of identical masses m interconnected by nonlinear springs. We assume that the springs have potential V (r), where r is the increase in distance between adjacent masses from the rest value at which the spring energy is minimum and its force (F = − dV dr ) is zero. If yn is the longitudinal displacement of
158
2 Nonlinear Model Equations and Variational Principles
the nth mass from its equilibrium position, it follows from the Newton second law of motion that d2 yn m 2 = V (yn+1 − yn ) − V (yn − yn−1 ). (2.3.33) dt With rn = (yn+1 − yn ), this gives an infinite set of differential equations dV (rn+1 ) dV (rn ) dV (rn ) dV (rn−1 ) m¨ rn = − − − , (2.3.34) drn+1 drn drn drn−1 where n ∈ N. In his celebrated paper, Toda (1967a, 1967b) investigated a mass–spring lattice system with an anharmonic potential in the form a (2.3.35) V (r) = e−br + br − 1 , a, b > 0. b With unit masses (m = 1), equation (2.3.34) reduces to the form (2.3.36) r¨n = a 2e−brn − e−brn+1 − e−brn−1 . This is known as the Toda lattice equation. In the limit as b → 0 with finite ab, equation (2.3.36) reduces to the linear differential-difference equation r¨n = ab(rn+1 − 2rn + rn−1 ). (2.3.37) √ This has solutions with a long wavelength velocity of ab lattice points per unit time. When b is not small, the Toda lattice equation (2.3.36) admits exact solitary wave solutions of the form (see Section 11.13) √
1 ab sinh κ , (2.3.38) rn = − log 1 + sinh2 κ sech2 κ n ± t b κ where the velocity of the lattice wave is expressed in terms of the amplitude parameter κ in the form sinh κ √ ab, (2.3.39) v= κ and the minus sign in (2.3.38) implies that the Toda lattice soliton (TLS) is a compression wave. As the amplitude of the TLS is reduced to zero (by letting sinh κ approach zero), it√reduces to a solution of the linear equation (2.3.37) traveling with velocity v = ab. We close this section by mentioning the Yang–Mills field equations which seem to be a useful model unifying electromagnetic and weak forces. They have solutions, called instantons, localized in space and time, which are interpreted as quantummechanical transitions between different states of a particle. Recently, it has been shown that the self-dual Yang–Mills equations are multidimensional integrable systems, and these equations admit reductions to well-known soliton equations in (1+1) dimensions, that is, the sine-Gordon, NLS, KdV, and Toda lattice equations (Ward 1984, 1985, 1986).
2.4 Variational Principles and the Euler–Lagrange Equations
159
2.4 Variational Principles and the Euler–Lagrange Equations Many physical systems are often characterized by their extremum (minimum, maximum, or saddle point) property of some associated physical quantity that appears as an integral in a given domain, known as a functional. Such a characterization is a variational principle leading to the Euler–Lagrange equation which optimizes the related functional. For example, light rays travel along a path from one point to another in a minimum time. The shortest distance between two points on a plane curve is a straight line. A physical system is in equilibrium if its potential energy is minimum. So the main problem is to optimize a physical quantity (time, distance, or energy) in most real-world problems. These problems belong to the subject of the calculus of variations. The classical Euler–Lagrange variational problem is to determine the extremum value of the functional b du , (2.4.1) I(u) = F (x, u, u ) dx, u = dx a with the boundary conditions u(a) = α
and
u(b) = β,
(2.4.2ab)
where α and β are given numbers and u(x) belongs to the class C 2 ([a, b]) of functions which have continuous derivatives up to the second order in a ≤ x ≤ b and the integrand F has continuous second derivatives with respect to all of its arguments. We assume that I(u) has an extremum at some u ∈ C 2 ([a, b]). Then we consider the set of all variations u + εv for finite u, and arbitrary v belonging to C 2 ([a, b]) such that v(a) = 0 = v(b). We next consider the variation δI of the functional I(u) δI = I(u + εv) − I(u) b
F (x, u + εv, u + εv ) − F (x, u, u ) dx = a
which, by the Taylor series expansion, b ∂F ∂F = F (x, u, u ) + ε v +v ∂u ∂u a 2 ∂F ε2 ∂F v + + v + · · · − F (x, u, u ) dx 2! ∂u ∂u b ∂F ∂F +v = ε v dx + O ε2 . ∂u ∂u a
(2.4.3)
Thus, a necessary condition for the functional I(u) to have an extremum (or for I(u) to be stationary) for an arbitrary ε is b ∂F ∂F 0 = δI = v dx, (2.4.4) +v ∂u ∂u a
160
2 Nonlinear Model Equations and Variational Principles
which, integrating the second term by parts, is b b d ∂F ∂F ∂F − v . = dx + v ∂u dx ∂u ∂u a a
(2.4.5)
Since v is arbitrary with v(a) = 0 = v(b), the last term of (2.4.5) vanishes and consequently, the integrand of the integral in (2.4.5) must vanish, that is, d ∂F ∂F = 0. (2.4.6) − ∂u dx ∂u This is called the Euler–Lagrange equation of the variational problem involving one independent variable. Using the result ∂F ∂ ∂F ∂ ∂F ∂ ∂F d (2.4.7) = dx + du + du , ∂u ∂x ∂u ∂u ∂u ∂u ∂u the Euler–Lagrange equations (2.4.6) can be written in the form Fu − Fxu − u Fuu − u Fu u = 0.
(2.4.8)
This is a second-order nonlinear ordinary differential equation for u provided Fu u = 0 and, hence, there are two arbitrary constants involved in the solution. However, when F does not depend explicitly on one of its variables x, u, or u , the Euler–Lagrange equation assumes a simplified form. Evidently, there are three possible cases: 1. If F = F (x, u), then (2.4.6) reduces to Fu (x, u) = 0, which is an algebraic equation. 2. If F = F (x, u ), then (2.4.6) becomes ∂F = const. ∂u
(2.4.9)
3. If F = F (u, u ), then (2.4.6) takes the form F − u Fu = const.
(2.4.10)
This follows from the fact that d d dF (F − u Fu ) = − u Fu − u Fu dx dx dx d = Fx + u Fu + u Fu − u Fu − u Fu dx d = u Fu − Fu = 0 by (2.4.6). dx The Euler–Lagrange variational problem involving two independent variables is to determine a function u(x, y) in a domain D ⊂ R2 satisfying the boundary conditions prescribed on the boundary ∂D of D and extremizing the functional
2.4 Variational Principles and the Euler–Lagrange Equations
I u(x, y) =
161
F (x, y, u, ux , uy ) dx dy,
(2.4.11)
D
where the function F is defined over the domain D and assumed to have continuous second-order partial derivatives. Similarly, for functionals depending on a function of two independent variables, the first variation δI of I is defined by δI = I(u + εv) − I(u). In view of Taylor’s expansion theorem, this reduces to
ε(vFu + vx Fp + vy Fq ) + O ε2 dx dy, δI =
(2.4.12)
(2.4.13)
D
where v = v(x, y) is assumed to vanish on ∂D and p = ux and q = uy . A necessary condition for the functional I to have an extremum is that the first variation of I vanishes, that is, (vFu + vx Fp + vy Fq ) dx dy 0 = δI = D ∂ ∂ Fp − Fq dx dy v Fu − = ∂x ∂y D ∂ ∂ Fp + Fq + (vx Fp + vy Fq ) dx dy + v ∂x ∂y D ∂ ∂ Fp − Fq dx dy = v Fu − ∂x ∂y D ∂ ∂ (vFp ) + (vFq ) dx dy. (2.4.14) + ∂y D ∂x We assume that the boundary curve ∂D has a piecewise, continuously moving tangent so that Green’s theorem can be applied to the second double integral in (2.4.14). Consequently, (2.4.14) reduces to ∂ ∂ v Fu − 0 = δI = Fp − Fq dx dy ∂x ∂y D v(Fp dy − Fq dx). (2.4.15) + ∂D
Since v = 0 on ∂D, the second integral in (2.4.15) vanishes. Moreover, since v is an arbitrary function, it follows that the integrand of the first integral in (2.4.15) must vanish. Thus, the function u(x, y) extremizing the functional defined by (2.4.11) satisfies the partial differential equation ∂F ∂ ∂ − Fp − Fq = 0. ∂u ∂x ∂y
(2.4.16)
162
2 Nonlinear Model Equations and Variational Principles
This is called the Euler–Lagrange equation for the variational problem involving two independent variables. The above variational formulation can readily be generalized for functionals depending on functions of three or more independent variables. Many physical problems require determining a function of several independent variables which will lead to an extremum of such functionals. Example 2.4.1. Find u(x, y) which extremizes the functional 2
ux + u2y dx dy, D ⊂ R2 . I u(x, y) =
(2.4.17)
D
The Euler–Lagrange equation with F = u2x + u2y = p2 + q 2 is ∂ ∂F ∂ ∂F + = 0, ∂x ∂p ∂y ∂q or uxx + uyy = 0. This is a two-dimensional Laplace equation. Similarly, the functional 2
ux + u2y + u2z dx dy dz, D ⊂ R3 , I u(x, y, z) =
(2.4.18)
(2.4.19)
D
will lead to the three-dimensional Laplace equation uxx + uyy + uzz = 0.
(2.4.20)
In this way, we can derive the n-dimensional Laplace equation ∇2 u = ux1 x1 + ux2 x2 + · · · + uxn xn = 0.
(2.4.21)
Example 2.4.2 (Plateau’s Problem). Find the surface S in the (x, y, z)-space of minimum area passing through a given plane curve C. The direction cosine of the angle between the z-axis and the normal to the surface 1 z = u(x, y) is (1 + u2x + u2y )− 2 . The projection of the element dS of the area of the 1 surface onto the (x, y)-plane is given by (1 + u2x + u2y )− 2 dS = dx dy. The area A of the surface S is given by 1 1 + u2x + u2y 2 dx dy, (2.4.22) A= D
where D is the area of the (x, y)-plane bounded by the curve C. 1 The Euler–Lagrange equation with F = (1 + p2 + q 2 ) 2 is given by ux uy ∂ ∂ + = 0. ∂x ∂y 1 + p2 + q 2 1 + p2 + q 2
(2.4.23)
2.4 Variational Principles and the Euler–Lagrange Equations
163
This is the equation of minimal surface, which reduces to the nonlinear elliptic partial differential equation (2.4.24) 1 + u2y uxx − 2ux uy uxy + 1 + u2x uyy = 0. Therefore, the desired function u(x, y) should be determined as the solution of the nonlinear Dirichlet problem for (2.4.24). This is difficult to solve. However, if the equation (2.4.23) is linearized around the zero solution, the square root term is replaced by one, and then the Laplace equation is obtained. Example 2.4.3 (Lagrange’s Equation in Mechanics). According to the Hamilton principle in mechanics, the first variation of the time integral of the Lagrangian L = L(qi , q˙i , t) of any dynamical system must be stationary, that is,
t2
0 = δI = δ
L(qi , q˙i , t) dt,
(2.4.25)
t1
where L = T − V is the difference between the kinetic energy, T , and the potential energy, V . In coordinate space, there are infinitely many possible paths joining any two positions. From all these paths, which start at a point A at time t1 and end at another point B at time t2 , nature selects the path qi = qi (t) for which δI = 0. Consequently, in this case, the Euler–Lagrange equation (2.4.6) reduces to ∂L d ∂L = 0, i = 1, 2, . . . , n. (2.4.26) − ∂qi dt ∂ q˙i In classical mechanics, these equations are universally known as the Lagrange equations of motion. The Hamilton function (or simply Hamiltonian) H is defined in terms of the generalized coordinates qi , generalized momentum pi = ∂∂L q˙i , and L by H=
n ∂L q ˙ (pi q˙i − L) = − L(qi , q˙i ) . i ∂ q˙i i=1 i=1
n
It readily follows that n n d ∂L d ∂L dH = 0. = (pi q˙i − L) = q˙i − dt dt i=1 dt ∂ q˙i ∂qi i=1
(2.4.27)
(2.4.28)
Thus, H is a constant, and hence, the Hamiltonian is the constant of motion. Example 2.4.4 (Hamilton’s Equations in Mechanics). To derive Hamilton equations of motion, we use the concepts of generalized momentum pi and generalized force Fi defined by pi =
∂L ∂ q˙i
and
(2.4.29a)
164
2 Nonlinear Model Equations and Variational Principles
Fi =
∂L . ∂qi
(2.4.29b)
Consequently, the Lagrange equations of motion (2.4.26) reduce to dpi ∂L = p˙ i . = ∂qi dt
(2.4.30)
In general, the Lagrangian L = L(qi , q˙i , t) is a function of qi , q˙i , and t where q˙i enters through the kinetic energy as a quadratic term. It then follows from the definition (2.4.27) of the Hamiltonian that H = H(pi , qi , t), and hence, its differential is ∂H ∂H ∂H dpi + dqi + dH = dt. (2.4.31) ∂pi ∂qi ∂t Differentiating (2.4.27) with respect to t gives d d ∂L d ∂L d ∂L dH = pi q˙i + qi − q˙i − , q˙i pi − dt dt dt ∂qi dt ∂ q˙i dt ∂t or equivalently, dH =
pi dq˙i +
q˙i dpi −
∂L ∂L ∂L dt, dqi − dq˙i − ∂qi ∂ q˙i ∂t
which, due to equation (2.4.29a), is =
q˙i dpi −
∂L ∂L dt. dqi − ∂qi ∂t
(2.4.32)
We next equate the coefficients of the two identical expressions (2.4.31) and (2.4.32) to obtain q˙i =
∂H , ∂pi
−
∂L ∂H = , ∂qi ∂qi
−
∂H ∂L = . ∂t ∂t
(2.4.33)
Using the Lagrange equations (2.4.30), the first two equations in (2.4.33) give q˙i =
∂H , ∂pi
p˙i = −
∂H . ∂qi
(2.4.34ab)
These are universally known as the Hamilton canonical equations of motion. Example 2.4.5 (Law of Conservation of Energy). The kinetic energy of a mechanical system described by a set of generalized coordinates qi is defined by T =
n n 1 i=1 j=1
2
aij q˙i q˙j ,
where aij are known functions of qi , and q˙i is the generalized velocity.
(2.4.35)
2.4 Variational Principles and the Euler–Lagrange Equations
165
In general, the potential energy V = V (qi , q˙i , t) is a function of qi , q˙i , and t. We assume here that V is independent of q˙i . For such a mechanical system, the Lagrangian is defined by L = T − V . Using the above definitions, the Hamilton principle states that, between any two points t1 and t2 , the actual motion takes place along the path qi = qi (t) such that the functional t2 t2 I qi (t) = L dt = (T − V ) dt, (2.4.36) t1
t1
is stationary (that is, the functional is an extremum). Or equivalently, the Hamilton principle can be stated as t2 (T − V ) dt = 0. (2.4.37) δI = δ t1
The integral I defined by (2.4.36) is often called the action integral of the system. Since the potential energy V does not depend on q˙i , it follows that ∂L ∂T = = aij q˙j , ∂ q˙i ∂ q˙i j=1 n
pi =
and the Hamiltonian H defined by (2.4.27) becomes n n n H= pi q˙i − L = q˙i aij q˙j − L = 2T − L = T + V. i=1
i=1
(2.4.38)
j=1
This proves that the Hamiltonian H is equal to the total energy. By (2.4.28), H is a constant, thus, the total energy of the system is constant. This is the celebrated law of conservation of energy. Example 2.4.6 (Motion of a Particle Under the Action of a Central Force). Consider the motion of a particle of mass m under the action of a central force −mF (r) where r is the distance of the particle from the center of force. The kinetic energy T is T =
1 1 mv 2 = m x˙ 2 + y˙ 2 , 2 2
which, in terms of polar coordinates, =
1 m 2
d (r cos θ) dt
2 +
d (r sin θ) dt
Since F = ∇V , the potential is given by V (r) = Then the Lagrangian L is given by
2 =
1 2 m r˙ + r2 θ˙2 . 2
(2.4.39)
r
F (r) dr.
(2.4.40)
166
2 Nonlinear Model Equations and Variational Principles
L=T −V =
1 m r˙ 2 + r2 θ˙2 − 2 2
r
F (r) dr .
Thus the Hamilton principle requires that the functional t2 t2 L dr = (T − V ) dt I(r, θ) = t1
(2.4.41)
(2.4.42)
t1
be stationary, that is, δI = 0. Consequently, the Euler–Lagrange equations are given by d d Lr − Lr˙ = 0 and Lθ − Lθ˙ = 0, (2.4.43) dt dt or equivalently, d 2 ˙ (2.4.44) r θ = 0. r¨ − rθ˙2 = −F (r) and dt These equations describe the planar motion of the particle. It follows immediately from the second equation of (2.4.44) that r2 θ˙ = const. = h.
(2.4.45)
In this case, rθ˙ represents the transverse velocity of the particle and mr2 θ˙ = mh is the constant angular momentum of the particle about the center of force. Introducing r = u1 , we find dr 1 du 1 du dθ du =− 2 =− · = −h , dt u dt u2 dθ dt dθ d du d2 r d2 u d2 u dθ r¨ = 2 = −h = −h 2 = −h2 u2 2 . dt dt dθ dθ dt dθ r˙ =
Substituting these into the first equation of (2.4.44) gives d2 u 1 1 . +u = 2 2F 2 dθ h u u
(2.4.46)
This is the differential equation of the central orbit, and it can be solved by standard methods. In particular, if the law of force is the attractive inverse square F (r) = μ/r2 so that the potential V (r) = −μ/r, the differential equation (2.4.46) becomes μ d2 u + u = 2. dθ2 h
(2.4.47)
If the particle is projected initially from the distance a with velocity v at an angle α that the direction of motion makes with the outward radius vector, then the constant h in (2.4.45) is h = av sin α. The angle φ between the tangent and radius vector of the orbit at any point is given by
2.4 Variational Principles and the Euler–Lagrange Equations
167
1 du d 1 1 dr =− =u . r dθ dθ u u dθ
cot φ =
The initial conditions at t = 0 are u=
du 1 = − cot α, dθ a
1 , a
when θ = 0.
(2.4.48)
The general solution of (2.4.47) is u=
μ 1 + e cos(θ + ε) , 2 h
(2.4.49)
where e and ε are constants to be determined by the initial data. Finally, the solution can be written as = 1 + e cos(θ + ε), r where =
h2 1 = (av sin α)2 . μ μ
(2.4.50)
(2.4.51)
This represents a conic section of semilatus rectum and eccentricity e with its axis inclined at the point of projection. The initial conditions (2.4.48) give = 1 + e cos ε, a so that
tan ε =
−a
− cot α = −e sin ε, a
cot α,
2 2 2 2 − 1 + 2 cot2 α = 1 − + 2 cosec2 α e = a a a a 1 1 2av 2 sin2 α + 2 a2 v4 sin2 α = 1− μ μ 2 2μ av sin α 2 = 1+ v − . μ a
2
(2.4.52)
Thus, the central orbit is an ellipse, parabola, or hyperbola accordingly as e < 1, = 1, or > 1, that is, v 2 < (2μ/a), = (2μ/a), or > (2μ/a). Example 2.4.7 (The Wave Equation of a Vibrating String). We assume that, initially, the string of length and line density ρ is stretched along the x-axis from x = 0 to x = . The string will be given a small lateral displacement, which is denoted by u(x, t) at each point along the x-axis at time t. The kinetic energy T of the string is given by
168
2 Nonlinear Model Equations and Variational Principles
1 2
T =
ρu2t dx,
(2.4.53)
0
and the potential energy is given by T =
T∗ 2
u2x dx,
(2.4.54)
0
where T ∗ is the constant tension of the string. According to the Hamilton principle t2 t2 0=δ L dt = δ (T − V ) dt t1 t2
=δ
t1
1 2
t1
ρu2t − T ∗ u2x dx dt.
(2.4.55)
0
In this case, L = 12 (ρu2t − T ∗ u2x ) which does not depend explicitly on x, t, or u, and hence, the Euler–Lagrange equation is given by ∂ ∂ (ρut ) − (T ∗ ux ) = 0, ∂t ∂x or utt − c2 uxx = 0,
(2.4.56)
where c2 = (T ∗ /ρ). This is the wave equation of the vibrating string. Example 2.4.8 (Two-Dimensional Wave Equation of Motion for Vibrating Membrane). We consider the motion of a vibrating membrane occupying the domain D under the action of a prescribed lateral force f (x, y, t) and subject to the homogeneous boundary conditions u = 0 on the boundary ∂D. The kinetic energy T and the potential energy V are given by 2 1 1 2 ux + u2y dx dy, (2.4.57) ut dx dy, V = μ T = ρ 2 2 D D where ρ is the surface density, μ is the elastic modulus of the membrane, and u = u(x, y, t) is the displacement function. The Lagrangian functional is of the form L dx dy, (2.4.58) L= D
where the Lagrangian density is given by L=
1 2 1 2 ρut − μ ux + u2y − uf (x, y, t). 2 2
(2.4.59)
According to the Hamilton principle, the first variation of the Lagrangian L must be stationary, that is,
2.4 Variational Principles and the Euler–Lagrange Equations
t2
0=δ
t2
L dt = δ t1
t1
D
1 2 1 2 ρut − μ ux + u2y − uf dx dy. 2 2
169
(2.4.60)
The Euler–Lagrange equation is ∂L ∂ ∂ ∂ − Lu − Lu − Lu = 0, ∂u ∂x x ∂y y ∂t t
(2.4.61)
−f + μ(uxx + uyy ) − ρutt = 0.
(2.4.62)
or equivalently, This leads to the nonhomogeneous wave equation μ∇2 u − ρutt = f (x, y, t).
(2.4.63)
This is the two-dimensional nonhomogeneous wave equation that can be solved with the initial conditions u(x, y, t = 0) = φ(x, y)
and ut (x, y, t = 0) = ψ(x, y)
at t = 0.
(2.4.64)
Example 2.4.9 (Three-Dimensional Nonhomogeneous Wave Equation). In threedimensional wave propagation in elastic media, the traveling waves exhibit various modes of vibration including longitudinal and transverse waves. To derive the appropriate equations of motion in continuous media, we need to extend the Hamilton principle by considering the displacement vector u = u(x, t). We use symmetric motion given by u = u(u1 , u2 , u3 ) where ui = ui (x1 , x2 , x3 , t), i = 1, 2, 3, and denote the particle velocity by ut = (u1,t , u2,t , u3,t ). Using this notation and tensor summation convention, the kinetic energy T and the potential energy V are given by T =
1 ρui,t ui,t 2
and
V =
1 μui,j ui,j , 2
(2.4.65)
∂ui i where ui,t = ∂u ∂t and ui,j = ∂xj . Introducing an external force term f (xi , t) so that the Lagrangian density is given by 1 1 L = ρui,t ui,t − μui,j ui,j − ui f (xi , t), (2.4.66) 2 2 the Lagrangian functional is of the form L= L dxj . (2.4.67) D
The generalized Hamilton principle for a three-dimensional continuum for various modes of wave propagation described by u(xj , t) takes the form t2 t2 dt L dxj = dt δL dxj . (2.4.68) 0 = δI(u) = δ t1
D
t1
D
This means that the function u = u(xj , t) makes the functional I(u) an extremum.
170
2 Nonlinear Model Equations and Variational Principles
Since L is a function of ui and ui,t , and the operator δ acts on the function ui and ui,t , we expand L to obtain δL =
∂L ∂L ∂L δui + δui,t + δui,j . ∂ui ∂ui,t ∂ui,j
(2.4.69)
We next substitute (2.4.69) into (2.4.68) and then integrate by parts with respect to t to obtain t2 t2 ∂L d ∂L δui,t dt. δui,t dt = − (2.4.70) t1 ∂ui,t t1 dt ∂ui,t ∂ and the δ variations in the integrals involving the spatial Interchanging ∂x j derivatives of ui , it turns out that ∂L ∂L ∂δui dxj , δui,j dxj = ∂xj D ∂ui,j D ∂ui,j
which, by integrating by parts, is ∂L δui − = ∂ui,t
D
d dxj
∂L δui dxj . ∂ui,j
(2.4.71)
Since ui vanishes at t1 and t2 , the integrated term also vanishes. Using (2.4.69)– (2.4.71) in (2.4.68) gives t2 ∂L d ∂L d ∂L dt δui − 0 = δI(u) = δ − dxj . ∂ui dt ∂ui,t dxj ∂ui,j t1 D (2.4.72) This is true only if the coefficients of each of the linearly independent displacements δui vanish. Consequently, (2.4.72) leads to the Euler–Lagrange equations of motion ∂L d d ∂L ∂L − − = 0, (2.4.73) ∂ui dt ∂ui,t dxj ∂ui,j where the summation over j is used. In particular, if the Lagrangian L is of the form (2.4.66), (2.4.73) gives the nonhomogeneous wave equations μ∇2 ui − ρui,tt = f (xi , t).
(2.4.74)
In the case of equilibrium, the Euler–Lagrange equations (2.4.74) reduce to the Poisson equation μ∇2 ui = f (xi , t). (2.4.75) We close this section by adding an important comment. Many equations in applied mathematics and mathematical physics can be derived from the Euler– Lagrange variational principle, the Hamilton principle, or from some appropriate variational principle.
2.6 The Variational Principle for Nonlinear Water Waves
171
2.5 The Variational Principle for Nonlinear Klein–Gordon Equations The nonlinear Klein–Gordon equation is utt − c2 uxx + V (u) = 0,
(2.5.1)
where V (u) is some nonlinear function of u chosen as the derivative of the potential energy V (u). The variational principle for equation (2.5.1) is given by δ L(u, ut , ux ) dt dx = 0, (2.5.2) where L is the associated Lagrangian density L(u, ut , ux ) =
1 2 ut − c2 u2x − V (u). 2
The Euler–Lagrange equation associated with (2.5.2) is ∂ ∂L ∂L ∂L ∂ − = 0, − ∂u ∂x ∂ux ∂t ∂ut which can be simplified to obtain the Klein–Gordon equation (2.5.1). We consider the variational principle δ L dx dt = 0,
(2.5.3)
(2.5.4)
(2.5.5)
with the Lagrangian L given by L≡
1 2 ut − c2 u2x − d2 u2 − γu4 , 2
(2.5.6)
where γ is a constant. The Euler–Lagrange equation associated with (2.5.5) gives the special case of the Klein–Gordon equation utt − c2 uxx + d2 u + 4γu3 = 0.
(2.5.7)
2.6 The Variational Principle for Nonlinear Water Waves In his pioneering work, Whitham (1965a, 1965b) first developed a general approach to linear and nonlinear dispersive waves using a Lagrangian. It is well known that most of the general ideas about dispersive waves have originated from the classical problems of water waves. So it is important to have a variational principle for water waves. Luke (1967) first explicitly formulated a variational principle for twodimensional water waves and showed that the basic equations and boundary and free surface conditions can be derived from the Hamilton principle.
172
2 Nonlinear Model Equations and Variational Principles
Fig. 2.1 A general surface gravity wave problem.
We now formulate the variational principle for three-dimensional water waves in the form δI = δ L dx dt = 0, (2.6.1) D
where the Lagrangian L is assumed to be equal to the pressure, so that η(x,t) 1 φt + (∇φ)2 + gz dz, L = −ρ 2 −h(x,y)
(2.6.2)
where D is an arbitrary region in the (x, t) space, ρ is the density of water, g is the gravitational acceleration, and φ(x, z, t) is the velocity potential in an unbounded fluid lying between the rigid bottom at z = −h(x, y) and the free surface z = η(x, y, t) as shown in Figure 2.1. The functions φ(x, z, t) and η(x, t) are allowed to vary subject to the restrictions δφ = 0 and δη = 0 at x1 , x2 , y1 , y2 , t1 , and t2 . Using the standard procedure in the calculus of variations, (2.6.1) becomes L dx dt 0 = −δ D ρ 1 φt + (∇φ)2 + gz = δη 2 D z=η η [φx δφx + φy δφy + φz δφz + δφt ] dz dx dt, (2.6.3) + −h
which, integrating the z-integral by parts, is 1 = δη φt + (∇φ)2 + gz 2 D z=η η η η ∂ ∂ ∂ + δφ dz + φx δφ dz + φy δφ dz ∂t −h ∂x −h ∂y −h η
(φxx + φyy + φzz )δφ dz − (ηt + ηx φx + ηy φy − φz )δφ z=η − −h
(2.6.4) + (φx hx + φy hy + φz )δφ z=−h dx dt.
2.7 The Euler Equation of Motion and Water Wave Problems
173
The second term within the square brackets integrates out to the boundaries ∂D of D and vanishes if δφ is chosen to be zero on ∂D. If we take δη = 0, [δφ]z=η = [δφ]z=−h = 0, since δφ is otherwise arbitrary; it turns out that ∇2 φ = 0,
−∞ < x, y < ∞, −h < z < η.
(2.6.5)
Since δη, [δφ]z=η , [δφ]z=−h may be given arbitrary independent values, it follows that 1 φt + (∇φ)2 + gη = 0 on z = η, 2 ηt + ηx φx + ηy φy − φz = 0 on z = η, φx hx + φy hy + φz = 0 on z = −h.
(2.6.6) (2.6.7) (2.6.8)
Thus, (2.6.5)–(2.6.8) represent the well-known nonlinear system of equations for classical water waves. Finally, this analysis is in perfect agreement with that of Luke (1967) and Whitham (1965a, 1965b, 1974) for two-dimensional waves on water of arbitrary but uniform depth h. It may be relevant to mention Zakharov’s (1968a, 1968b) Hamiltonian formulation. The Hamiltonian is η 1 ∞ 2 H= gη 2 + (∇φ) dz dx. (2.6.9) 2 −∞ −h On the other hand, Benjamin and Olver (1982) have described the Hamiltonian structure, symmetrics, and conservation laws for water waves. Olver (1984a, 1984b) has discussed Hamiltonian and non-Hamiltonian models for water waves, and Hamiltonian perturbation theory and nonlinear water waves.
2.7 The Euler Equation of Motion and Water Wave Problems The Euler equation of motion and the equation of continuity have provided the fundamental basis of the study of modern theories of water waves, which are the most common observable phenomena in nature. Water wave motions are of great importance as they range from waves generated by wind or solar heating at the surface of oceans to flood waves in rivers, from waves caused by a moving ship in a channel to tsunami (tidal waves) generated by earthquakes, and from solitary waves on the surface of a channel generated by a disturbance to waves generated by underwater explosions, to mention only a few. Making reference to Debnath’s book (1994), Nonlinear Water Waves, the Euler equation of motion in an inviscid and incompressible fluid of constant density ρ under the action of body force F = (0, 0, −g) where g is the constant acceleration of gravity and the equation of continuity are given by 1 Du = − ∇p + F, Dt ρ
(2.7.1)
174
2 Nonlinear Model Equations and Variational Principles
∇·u =
∂u ∂v ∂w + + = 0, ∂x ∂y ∂z
(2.7.2)
where x = (x, y, z) is the rectangular Cartesian coordinates and u = (u, v, w) is the velocity vector, p is the pressure field, and ∂ ∂ ∂ ∂ ∂ D = +u·∇= +u +v +w . Dt ∂t ∂t ∂x ∂y ∂z
(2.7.3)
These equations constitute a closed system of four nonlinear partial differential equations for four unknowns u, v, w, and p. So these equations with appropriate initial and boundary conditions are sufficient to determine the velocity field u and pressure p uniquely. In the study of water waves, the body force is always the acceleration due to gravity, that is, F = (0, 0, −g). It is convenient to write the three components of the Euler equation in the form 1 ∂p Du =− , Dt ρ ∂x 1 ∂p Dv =− , Dt ρ ∂y Dw 1 ∂p =− − g, Dt ρ ∂z
(2.7.4) (2.7.5) (2.7.6)
and the continuity equation (2.7.2). In cylindrical polar coordinates x = (r, θ, z) with the velocity vector u = (u, v, w), the Euler equations and the continuity equation are given by Du v 2 − Dt r uv Dv + Dt r Dw Dt 1 ∂v ∂w 1 ∂ (ru) + + r ∂r r ∂θ ∂z where
1 ∂p , ρ ∂r 1 1 ∂p =− , ρ r ∂θ 1 ∂p =− − g, ρ ∂z
=−
= 0,
(2.7.7) (2.7.8) (2.7.9) (2.7.10)
∂ ∂ v ∂ ∂ D = +u + +w . (2.7.11) Dt ∂t ∂r r ∂θ ∂z One of the fundamental properties of a fluid flow is called the vorticity, which is defined by the curl of the velocity field so that ω = curl u = ∇ × u. The vorticity vector ω measures the local spin or rotation of individual fluid particles. Evidently, fluid flows in which ω = 0 are called irrotational. In the real world, fluid flows are hardly irrotational anywhere; however, for many flows the vorticity is very small almost everywhere and the fluid motion may be treated as irrotational. In problems
2.7 The Euler Equation of Motion and Water Wave Problems
175
of water waves, the motion of fluid is considered unsteady and irrotational which implies that the vorticity ω = curl u = ∇ × u = 0. So there exists a single-valued velocity potential φ so that u = ∇φ. The continuity equation (2.7.2) then reduces to the Laplace equation ∇2 φ =
∂ 2φ ∂2φ ∂2φ + 2 + 2 = 0. ∂x2 ∂y ∂z
(2.7.12)
Using the vector identity, (u · ∇)u = 12 ∇u2 − u × ω combined with ω = 0 and u = ∇φ, the Euler equation (2.7.1) may be written in the form 1 p ∇ φt + (∇φ)2 + + gz = 0. (2.7.13) 2 ρ This can be integrated with respect to the space variables to obtain the equation 1 p φt + (∇φ)2 + + gz = c(t), 2 ρ
(2.7.14)
where c(t) is an arbitrary function of time only (∇c = 0) determined by the pressure imposed at the boundaries of the fluid flow. Since only the pressure gradient affects the flow, a function of t alone added to the pressure field p has virtually no effect on the motion. So, without loss of generality, we can set c(t) ≡ 0 in (2.7.14). Consequently, equation (2.7.14) becomes 1 p φt + (∇φ)2 + + gz = 0. 2 ρ
(2.7.15)
This equation is known as Bernoulli’s equation (or the pressure equation), which completely determines the pressure in terms of the velocity potential φ. Thus, the Laplace equation (2.7.12) and (2.7.15) are used to determine φ and p, and hence the velocity components u, v, w, and the pressure p. In cylindrical polar coordinates x = (r, θ, z) with the velocity field u = (φr , 1r φθ , φz ), the Laplace equation becomes ∇2 φ =
1 ∂ 1 ∂2φ ∂2φ (rφr ) + 2 2 + 2 = 0. r ∂r r ∂θ ∂z
(2.7.16)
We assume that the fluid occupies the region −h ≤ z ≤ 0 with the plane z = −h as the rigid bottom boundary and the plane z = 0 as the upper (free surface) boundary in the undisturbed state. We suppose that the upper boundary is the surface exposed to a constant atmospheric pressure pa . Since the free surface is exposed to the constant atmospheric pressure pa , we have p = pa on this surface. After the motion is set up, we denote this surface by S with the equation z = η(x, y, t) where η is an unknown function of x, y, and t that tends to zero as t → 0. The function η(x, y, t) is referred to as the free surface elevation. The rate of change of η, following a fluid particle, is equal to the vertical component of ∇φ at the surface, that is,
176
2 Nonlinear Model Equations and Variational Principles
ηt + u · ∇η = φz
on z = η.
Or equivalently, this free surface condition reads as ηt + φx ηx + φy ηy − φz = 0
on z = η.
(2.7.17)
This is called the kinematic free surface condition. Since p = pa on S, after absorbing pρa and c(t) into φt , equation (2.7.14) can be rewritten as ∂φ 1 + (∇φ)2 + gz = 0 on S for t ≥ 0. (2.7.18) ∂t 2 Since S is a free boundary surface, it contains the same fluid particles for all times, that is, S is a material surface. Hence, it follows from (2.7.18) that D ∂φ 1 2 + (∇φ) + gz = 0 on S for t ≥ 0. Dt ∂t 2 Or equivalently, on S for t ≥ 0, ∂φ 1 ∂ 2 + ∇φ · ∇ + (∇φ) + gz ∂t ∂t 2 1 = φtt + 2∇φ · ∇(φt ) + ∇φ · ∇(∇φ)2 + gφz = 0. 2
(2.7.19)
Since the bottom boundary z = −h is a rigid solid surface at rest, the condition to be satisfied at this boundary is ∂φ = 0 on z = −h, t ≥ 0. ∂z
(2.7.20)
Thus, Laplace’s equation (2.7.12) together with the free surface boundary conditions (2.7.17), (2.7.19) and the bottom boundary condition (2.7.20) determine the velocity potential φ and the free surface elevation η. Because of the presence of the nonlinear terms in the free surface boundary conditions (2.7.17) and (2.7.19), the determination of φ and η in the general case is a difficult task. We restrict our discussion to two particular cases because of the great importance of water wave motions. Example 2.7.1 (Small Amplitude Water Waves). We consider plane waves propagating in the x-direction whose amplitude varies in the z-direction with the gravitational force as the only body force. We first consider the case where the motion is linear so that nonlinear terms in velocity components may be neglected. In this case, no distinction is made between the initial and the current states of the free surface boundary, and the boundary conditions (2.7.18) and (2.7.19) are given in the linearized forms φt + gη = 0 on z = 0, t > 0, φtt + gφz = 0 on z = 0, t > 0.
(2.7.21) (2.7.22)
2.7 The Euler Equation of Motion and Water Wave Problems
177
These conditions yield ηt = φ z
on z = 0, t > 0.
(2.7.23)
For a plane wave propagating in the x-direction with frequency ω and wavenumber k, we seek a solution for φ(x, z, t) in the form
φ = Φ(z) exp i(ωt − kx) , (2.7.24) where Φ(z) is a function to be determined. Substituting (2.7.24) in the Laplace equation (2.7.12) with no y dependence gives an equation for Φ as (2.7.25) Φzz = k 2 Φ. The general solution of this equation is Φ(z) = Aekz + Be−kz ,
(2.7.26)
where A and B are arbitrary constants. Using the boundary condition (2.7.20), we find A exp(−kh) = B exp(kh) so that the solution (2.7.26) takes the form Φ = C cosh k(z + h),
(2.7.27)
where C = 2A exp(−kh) = 2B exp(kh) is an arbitrary constant so that the solution (2.7.24) becomes
φ = C cosh k(z + h) exp i(ωt − kx) . (2.7.28) Using (2.7.28) in (2.7.21) yields
η = a exp i(ωt − kx) ,
(2.7.29)
where a = (Cω/ig) cosh kh = max |η| is the amplitude. Thus, the solution (2.7.28) assumes the final form
iag cosh k(z + h) exp i(ωt − kx) . (2.7.30) φ= ω cosh kh Substituting (2.7.30) into (2.7.22) gives the following dispersion relation between the frequency and wavenumber: ω 2 = gk tanh kh.
(2.7.31)
Thus, the phase velocity, cp = ( ωk ), can be obtained from (2.7.31) as c2p =
ω2 g = tanh(kh). 2 k k
(2.7.32)
This result shows that the phase velocity cp depends on the wavenumber k, depth h, and the gravity g. Hence, water waves are dispersive in nature. This means that, as
178
2 Nonlinear Model Equations and Variational Principles
the time passes, the waves would disperse (spread out) into different groups such that each group would consist of waves having approximately the same wavelength. The quantity dω dk represents the velocity of such a group in the direction of propagation and is called the group velocity, cg . It follows from (2.7.31) that g dω cg = tanh kh + kh sec h2 kh , = (2.7.33) dk 2ω which, by using (2.7.32), is =
2kh 1 . cp 1 + 2 sinh 2kh
(2.7.34)
Evidently, the group velocity is different from the phase velocity. In the case where wavelength 2π/k is large compared with the depth h, such waves are called long waves (or shallow water waves), kh 1 so that tanh kh ≈ kh, and hence, sinh 2kh ≈ 2kh. In such a situation, results (2.7.32) and (2.7.34) give (2.7.35) cg = cp ≈ gh = c. Thus, shallow water waves are nondispersive and their speed varies as the square root of the depth. In the other limiting case, where the wavelength is very small compared with the depth, such waves are called short waves (or deep water waves), kh 1. In the limit kh → ∞, [cosh k(z + h)/ cosh kh] → exp(kz), and the corresponding solutions for φ and η become
iag exp kz + i(ωt − kx) φ = Re ω ag exp(kz) sin(kx − ωt), (2.7.36) = ω
η = Re a exp i(ωt − kx) = a cos(ωt − kx). (2.7.37) In the limit kh → ∞, tanh kh → 1 so that the dispersion relation becomes ω 2 = gk.
(2.7.38)
Consequently, 12 12 g gλ cp = = , k kπ 1 cg = cp . 2
(2.7.39) (2.7.40)
Evidently, deep water waves are dispersive and the phase velocity is proportional to the square root of their wavelength. Also the group velocity is equal to one-half of the phase velocity.
2.7 The Euler Equation of Motion and Water Wave Problems
179
Example 2.7.2 (The Stokes’ Waves or Nonlinear Finite Amplitude Waves). We consider the Stokes’ waves where the motion is nonlinear and the amplitude of the waves is not small. We recall Bernoulli’s equation (2.7.18) and (2.7.19) and write them for ready reference in the form 1 1 2 , (2.7.41) η = − φt + (∇φ) g 2 z=η
1 [φtt + gφz ]z=η + 2[∇φ · ∇φt ]z=η + ∇φ · ∇(∇φ)2 z=η = 0. (2.7.42) 2 A systematic procedure can be employed to rewrite these boundary conditions by using Taylor’s series expansions of the potential φ and its derivatives in the form 1 (2.7.43) φ(x, y, z = η, t) = [φ]z=0 + η[φz ]z=0 + η 2 [φzz ]z=0 + · · · , 2 1 φz (x, y, z = η, t) = [φz ]z=0 + η[φzz ]z=0 + η 2 [φzzz ]z=0 + · · · . (2.7.44) 2 Substituting these and similar Taylor’s expansions into (2.7.41) gives
1 1 1 1 2 2 + η − φt + (∇φ) + ··· η = − φt + (∇φ) g 2 g 2 z z=0 z=0 1 1 = − φt + (∇φ)2 g 2 z=0 1 1 1 2 2 + 2 φt + (∇φ) + ··· φt + (∇φ) g 2 2 z z=0 1 1 1 = − φt + (∇φ)2 − φt φzt + O φ3 . (2.7.45) g 2 g z=0 Similarly, condition (2.7.42) gives
1
[φtt + gφz ]z=0 + η (φtt + gφz )z + η 2 (φtt + gφz )zz z=0 2
+ · · · + 2[∇φ · ∇φt ]z=0 + 2η {∇φ · ∇φt }z z=0 + η 2 {∇φ · ∇φt }zz z=0
1 1 + ··· + ∇φ · ∇(∇φ)2 z=0 + η ∇φ · ∇(∇φ)2 z z=0 2 2
1 + η 2 ∇φ · ∇(∇φ)2 zz z=0 + · · · = 0. (2.7.46) 4 We substitute (2.7.45) for η into (2.7.46) to obtain
1 1 1 2 (φtt + gφz )z z=0 [φtt + gφz ]z=0 − φt + (∇φ) − φt φzt g 2 g z=0 2
1 1 1 + 2 φt + (∇φ)2 − φt φzt (φtt + gφz )zz z=0 2g 2 g z=0
180
2 Nonlinear Model Equations and Variational Principles
2 1 1 (∇φ · ∇φt )z z=0 + 2 (∇φ) · ∇φt z=0 − φt + (∇φ)2 − φt φzt g 2 g z=0
1 2 1 1 2 2 + ∇φ · ∇(∇φ) z=0 − φt + (∇φ) − φt φzt 2 g 2 g z=0
2 × ∇φ · ∇(∇φ) z z=0 = 0. (2.7.47) The first-, second-, and third-order boundary conditions on z = 0 are respectively given by (2.7.48) (φtt + gφz ) = 0 + O φ2 , 3 1 (φtt + gφz ) + 2[∇φ · ∇φt ] − φt (φtt + gφz )z = 0 + O φ , (2.7.49) g
1 (φtt + gφz ) + 2[∇φ · ∇φt ] + ∇φ · ∇(∇φ)2 2
1 1 2 1 1 ∇φ − φt φzt [φtt + gφz ]z − φt φtt + gφz + 2(∇φ · ∇φt ) z − g g 2 g
1 + 2 [φt ]2 (φtt + gφz )zz = 0 + O φ4 , (2.7.50) 2g where O (·) indicates the order of magnitude of the neglected terms. These results can be determined by the third-order expansion of plane progressive surface waves. As indicated before, the first-order plane wave potential φ in deep water is given by (2.7.36). Direct substitution of the first-order velocity potential (2.7.36) in the second-order boundary condition (2.7.49) reveals that the second-order terms in (2.7.49) vanish. Thus, the first-order potential is a solution of the second-order boundary-value problem, and we can state that ga kz e sin(kx − ωt) + O a3 . (2.7.51) φ= ω Substitution of this result into (2.7.45) leads to the second-order result for η in the form 1 η = a cos(kx − ωt) − ka2 + ka2 cos2 (kx − ωt) + · · · 2 1 = a cos(kx − ωt) + ka2 cos 2(kx − ωt) + · · · . 2
(2.7.52)
The second term in (2.7.52), which represents the second-order correction to the surface profile, is positive at the crests kx − ωt = 0, 2π, 4π, . . . , and negative at the troughs kx − ωt = π, 3π, 5π, . . . . But the crests are steeper, and the troughs flatter as a result of the nonlinear effect. The notable feature of solution (2.7.52) is that the wave profile is no longer sinusoidal. The actual shape of the wave profile is a curve known as a trochoid (see Figure 2.2), whose crests are steeper and troughs are flatter than those of the sinusoidal wave. Substituting the wave potential (2.7.51) in the third-order boundary condition (2.7.50) reveals that all nonlinear terms vanish identically except one term, ( 12 )∇φ ·
2.7 The Euler Equation of Motion and Water Wave Problems
181
Fig. 2.2 The surface wave profile.
∇(∇φ)2 . Thus the boundary condition for the third-order plane wave solution is given by 1 (2.7.53) φtt + gφz + ∇φ · ∇(∇φ)2 = 0 + O φ4 . 2 If the first-order solution (2.7.51) is substituted into the third-order boundary condition on z = 0, the dispersion relation with second-order effect is obtained in the form (2.7.54) ω2 = gk 1 + a2 k 2 + O k3 a3 . Note that this relation involves the amplitude in addition to frequency and wavenumber. This is called the nonlinear dispersion relation and it can be expressed in terms of the phase velocity as cp =
ω = k
12 12 1 g 1 g 1 + a2 k 2 . 1 + k 2 a2 2 ≈ k k 2
(2.7.55)
Thus the phase velocity depends on the wave amplitude, and waves of large amplitude travel faster than smaller ones. The dependence of cp on amplitude is known as the amplitude dispersion in contrast to the frequency dispersion as given by (2.7.38). It may be noted that Stokes’ results (2.7.52), (2.7.54), and (2.7.55) can easily be approximated further to obtain solutions for long waves (or shallow water) and for short waves (or deep water). We conclude this example by discussing the phenomenon of breaking of water waves, which is one of the most common observable phenomena on an ocean beach. A wave coming from deep ocean changes shape as it moves across a shallow beach. Its amplitude and wavelength also are modified. The wavetrain is very smooth some distance offshore, but as it moves inshore, the front of the wave steepens noticeably until, finally, it breaks. After breaking, waves continue to move inshore as a series of bores or hydraulic jumps, whose energy is gradually dissipated by means of the water turbulence. Of the phenomena common to waves on beaches, breaking is the physically most significant and mathematically least known. In fact, it is one of the most intriguing longstanding problems of water waves theory. For waves of small amplitude in deep water, the maximum particle velocity is v = aω = ack. But the basic assumption of small amplitude theory implies that vc = ak 1. Therefore, wave breaking can never be predicted by the small amplitude wave theory. That possibility arises only in the theory of finite amplitude waves. It is
182
2 Nonlinear Model Equations and Variational Principles
Fig. 2.3 The steepest wave profile.
to be noted that the Stokes’ expansions are limited to relatively small amplitude and cannot predict the wavetrain of maximum height at which the crests are found to be very sharp. For a wave profile of constant shape moving at a uniform velocity, it can be shown that the maximum total crest angle as the wave begins to break is 120◦ ; see Figure 2.3. The upshot of the Stokes’ analysis reveals that the inclusion of higher-order terms in the representation of the surface wave profile distorts its shape away from the linear sinusoidal curve. The effects of nonlinearity are likely to make crests narrower (sharper) and the troughs flatter as depicted in Figure 2.7 of Debnath (1994). The resulting wave profile more accurately portrays the water waves that are observed in nature. Finally, the sharp crest angle of 120◦ was first found by Stokes. On the other hand, in 1865, Rankine conjectured that there exists a wave of extreme height. In a moving reference frame, the Euler equations are Galilean invariant, and the Bernoulli equation (2.7.14) on the free surface of water with ρ = 1 becomes 1 |∇φ|2 + gz = E. 2 Thus, this equation represents the conservation of local energy, where the first term is the kinetic energy of the fluid and the second term is the potential energy due to gravity. For the wave of maximum height, E = gzmax , where zmax is the maximum height of the fluid. Thus, the velocity is zero at the maximum height so that there will be a stagnation point in the fluid flow. Rankine conjectured that a cusp is developed at the peak of the free surface with a vertical slope so that the angle subtended at the peak is 120◦ as also conjectured by Stokes (1847). Toland (1978) and Amick et al. (1982) have proved rigorously the existence of a wave of greatest height and the Stokes’ conjecture for the wave of extreme form. However, Toland (1978) also proved that if the singularity at the peak is not a cusp, that is, if there is no vertical slope at the peak of the free surface, then the Stokes’ remarkable conjecture of the crest angle of 120◦ is true. Subsequently, Amick et al. (1982) confirmed that the singularity at the peak is not a cusp. Therefore, the full Euler equations exhibit singularities, and there is a limiting amplitude to the periodic waves. We next formulate the modern mathematical theory of nonlinear water waves. It is convenient to take the free surface elevation above the undisturbed mean depth h as z = η(x, y, t) so that the free surface of water is at z = H = h + η and the horizontal rigid bottom is at z = 0 where the z-axis is vertical positive upwards.
2.7 The Euler Equation of Motion and Water Wave Problems
183
It is also convenient to introduce nondimensional flow variables based on a typical horizontal length scale (which may be wavelength λ), typical vertical length √ scale h, typical horizontal velocity scale, c = gh (shallow water wave speed), ), and the typical pressure typical time scale ( c ), typical vertical velocity scale ( hc scale ρc2 . Using asterisks to denote nondimensional flow variables, we write ∗ ∗ ∗ ∗ z = hz , t= (2.7.56) (x, y) = x , y , t , c hc w= (u, v) = c u∗ , v ∗ , w∗ , and p = ρc2 p∗ . (2.7.57) We next introduce two fundamental parameters δ and ε defined by δ=
h2 2
and
ε=
a , h
(2.7.58)
where δ is called the long wavelength (or shallowness) parameter, ε is called the amplitude parameter, and a is the typical wave amplitude. These two parameters play a crucial role in the modern theory of water waves. In terms of the amplitude parameter, the free surface at z = H = h + η and the bottom boundary surface at z = 0 of the fluid can be written as the nondimensional form z = 1 + εη and z = 0, (2.7.59) where ( ηa ) is replaced by the nondimensional value η. The variable pressure field P representing the deviation from the hydrostatic pressure gρ(h − z) is given by p = pa + gρ(z − h) + gρhP,
(2.7.60)
where pa is the constant atmospheric pressure and the scale gρh of pressure is based on the pressure at the depth z = h. In terms of the nondimensional variables, the Euler equations (2.7.4)–(2.7.6) and the continuity equation (2.7.2) can be written in the form, dropping the asterisks and replacing P by p, ∂p Dv ∂p Du =− , =− , Dt ∂x Dt ∂y ∂u ∂v ∂w + + = 0. ∂x ∂y ∂z
δ
Dw ∂p =− , Dt ∂z
(2.7.61) (2.7.62)
The kinematic free surface and the dynamic free surface conditions (see Debnath 1994, pp. 6–7) are expressed in the nondimensional form, dropping the asterisks, w = ε(ηt + uηx + vηy ) on z = 1 + εη, p = εη on z = 1 + εη. The bottom boundary condition is
(2.7.63) (2.7.64)
184
2 Nonlinear Model Equations and Variational Principles
w = 0 on z = 0.
(2.7.65)
It follows from the free surface conditions that both w and p are proportional to the amplitude parameter ε. In the limit as ε → 0, both w and p tend to zero, indicating that there is no disturbance at the free surface. Consistent with the governing equations and the boundary conditions, we introduce a set of scaled flow variables (u, v, w, p) → ε(u, v, w, p)
(2.7.66)
so that the governing equations (2.7.61) and (2.7.62) and the boundary conditions (2.7.63)–(2.7.65) become ∂p Dv ∂p Du =− , =− , Dt ∂x Dt ∂y ∂u ∂v ∂w + + = 0, ∂x ∂y ∂z w = ηt + ε(uηx + vηy ), p = η
δ
Dw ∂p =− , Dt ∂z
(2.7.68) on z = 1 + εη,
w = 0 on z = 0, where
(2.7.67)
(2.7.69) (2.7.70)
∂ ∂ ∂ ∂ D ≡ +ε u +v +w , Dt ∂t ∂x ∂y ∂z
(2.7.71)
and parameters δ and ε are given by (2.7.58). In general, there are two most commonly adopted and useful approximations: (i) ε → 0, that is, small amplitude water waves governed by the linearized equations, and (ii) δ → 0, that is, shallow water wave equations (or long water waves). These approximate models and their solutions constitute the classical theory of water waves (see Debnath 1994). In the first case, the linearized equations of water waves are obtained from (2.7.67)–(2.7.71) in the limit as ε → 0 in the form ∂p ∂v ∂p ∂u =− , =− , ∂t ∂x ∂y ∂y ∂u ∂v ∂w + + = 0, ∂x ∂y ∂z w = ηt and p = η on z = 1,
δ
∂w ∂p =− , ∂t ∂z
(2.7.72) (2.7.73) (2.7.74) (2.7.75)
w = 0 on z = 0.
In the second√case, the shallow water equations (long water waves) are described in the sense that δ = ( h ) is small so that δ → 0 with fixed amplitude parameter ε. Consequently, the governing equations and the boundary conditions are obtained from (2.7.67)–(2.7.71) in the limit as δ → 0 in the form Du ∂p =− , Dt ∂x
Dv ∂p =− , Dt ∂y
∂p = 0, ∂z
(2.7.76)
2.7 The Euler Equation of Motion and Water Wave Problems
∂u ∂v ∂w + + = 0, ∂x ∂y ∂z w = ηt + ε(uηx + vηy ) w = 0 on z = 0,
185
(2.7.77) and
p=η
on z = 1 + εη,
(2.7.78) (2.7.79)
D where Dt is given by (2.7.71). Finally, equations that describe small amplitude waves ε → 0 and long waves (δ → 0) are obviously consistent with (2.7.72)–(2.7.75) for the first case, and also with equations (2.7.76)–(2.7.79) with
∂p = 0; ∂z
p=η
on z = 1
(2.7.80)
or (2.7.76)–(2.7.79) with ε → 0. The solutions of these various approximate governing equations describe the classical water waves (see Debnath 1994). Example 2.7.3 (Solution of a Linearized Water Wave Problem). We consider the propagation of a plane harmonic water wave in the x-direction in a fluid of constant depth. With no y-dependence, the governing equations and the boundary conditions are obtained from (2.7.72)–(2.7.75) in the form ut = −px , δwt = −pz , ux + wz = 0, w = ηt and p = η on z = 1, w = 0 on z = 0. We assume a plane wave solution in the form
(u, w, p) = u∗ (z), w∗ (z), p∗ (z) exp i(ωt − kx) . The free surface elevation is given by
η(x, t) = a exp i(ωt − kx) + c.c.,
(2.7.81) (2.7.82) (2.7.83)
(2.7.84)
(2.7.85)
where a is a constant wave amplitude and c.c. denotes the complex conjugate. Obviously, (2.7.85) represents a plane harmonic wave whose initial form at t = 0 is given by η(x, 0). Substituting the solution (2.7.84) into (2.7.81) gives, dropping asterisks, u=
k p, ω
p = −iωδw,
w = iku,
(2.7.86)
where the prime denotes the derivative with respect to z. It readily follows from (2.7.86) that w = iku =
ik 2 2 p = δk w. ω
Thus, the general solution of (2.7.87) is
(2.7.87)
186
2 Nonlinear Model Equations and Variational Principles
√ √ w = A exp δkz + B exp − δkz ,
(2.7.88)
where A and B are arbitrary constants to be determined from (2.7.82), (2.7.83) which give w(1) = iωa, p(1) = a, w(0) = 0. (2.7.89) Consequently, the solution (2.7.88) assumes the final form √ sinh( δkz) √ . w(z) = Re(iωa) sinh( δk)
(2.7.90)
It follows from the boundary conditions that 2 √ ω ω aω √ a = p(1) = u(1) = 2 w (1) = δ coth k δ . k ik k
(2.7.91)
This leads to the dispersion relation √ k (2.7.92) tanh k δ , ω2 = √ δ √ √ where δ = hλ−1 which is equal to dimensional (physical) quantity ( h ) and k δ is equal to dimensional quantity kh. As before, the dispersion relation determines the frequency ω = ω(k) and the phase velocity 2 √ −1 √ ω = k δ tanh k δ . (2.7.93) c2p = k The group velocity cg is given by cg =
√ √
√ 1 dω √ tanh k δ + k δ sec h2 k δ = dk 2ω δ
which, by (2.7.93), is
√ 2k δ 1 √ = cp 1 + . (2.7.94) 2 sinh(2k δ) √ √ √ In the case of shallow water waves, k δ → 0 so that tanh(k δ) ≈ k δ. Hence, results (2.7.93), (2.7.94) lead to cp = cg = 1. Both the phase and group velocities are independent of the wavelength. So, the shallow water waves are nondispersive. In terms of the dimensional variables, the phase velocity is (2.7.95) cp = ±c = ± gh. This confirms the choice of the √ before. √ velocity scale c adopted For deep water waves, k δ → ∞ so that tanh(k δ) → 1. Consequently, k ω2 = √ , δ
√ −1 c2p = k δ ,
and cg =
1 cp . 2
(2.7.96)
2.7 The Euler Equation of Motion and Water Wave Problems
187
Example 2.7.4 (Small Amplitude Gravity-Capillary Surface Waves on Water of depth h). The governing equation for the two-dimensional linearized gravitycapillary surface waves on water of constant depth h with the free surface at z = 0 are given by (2.7.97) ∇2 φ = φxx + φzz = 0, −h ≤ z < 0, t > 0, where φ(x, z, t) is the velocity potential. Representing the free surface elevation function by η = η(x, t), the linearized kinematic and dynamic free surface conditions are ηt − φz = 0 on z = 0, t > 0, T φt + gη − ηxx = 0 on z = 0, t > 0, ρ
(2.7.98) (2.7.99)
where g is the acceleration of gravity and T is the surface tension, and ρ is the constant density of water. The boundary condition at the horizontal rigid bottom at z = −h φz = 0 at z = −h.
(2.7.100)
We seek the same solution (2.7.28) for φ(z, x, t) and (2.7.29) for η(x, t) so that φ(x, z, t) assumes the same form (2.7.30). In view of (2.7.98), (2.7.99) reduces to the form T (2.7.101) φtt + gφz − φzxx = 0 on z = 0, t > 0. ρ Substituting (2.7.30) into (2.7.101) gives the dispersion relation for the gravitycapillary waves in the form T k2 2 tanh kh. (2.7.102) ω = gk 1 + ρg Or equivalently, this gives the phase velocity cp of the surface gravity-capillary waves on water of finite depth h T ω2 g + k tanh kh. (2.7.103) c2p = 2 = k k ρ It can easily be recognized that result (2.7.102) or (2.7.103) is exactly the same as result (2.7.31) or (2.7.32) with g replaced by (g + ρ−1 T k 2 ). This means that all the properties of gravity-capillary waves can be described correctly when this replacement is made in the results of pure gravity waves. Introducing the Froude number F and the Bond number τ by cp F =√ gh
and
τ=
T , gρh2
the dimensionless form of the dispersion relation (2.7.103) is
(2.7.104)
188
2 Nonlinear Model Equations and Variational Principles
Fig. 2.4 The square of the Froude number, F 2 , against (kh)−1 .
2
F =
1 + τ kh tanh kh. kh
(2.7.105)
λ The square of the Froude number is plotted against (kh)−1 = ( 2πh ) for four values 1 of the Bond number τ = 1.3, 3 , 0.1, and 0.05 in Figure 2.4. This figure shows that F 2 decreases monotonically with (kh)−1 when τ > 13 , and it has a minimum for τ < 13 . As kh = 2π( hλ ) → 0 (or λh → ∞), F → 1. In case of water of infinite depth (kh → ∞, tanh kh → 1), (2.7.102) or (2.7.103) leads to Tk T k2 g 2 2 or cp = . (2.7.106) + ω = gk 1 + gρ k ρ
Thus, for pure surface gravity waves (T = 0, g = 0) in deep water, (2.7.106) reduces to (2.7.38). Similarly, for pure surface capillary waves (g = 0, T = 0), the dispersion relation is T k3 Tk or c2p = . (2.7.107) ω2 = ρ ρ It is convenient to write (2.7.106) as ω2 = gk(1 + T ∗ )
or
c2p =
g (1 + T ∗ ), k
(2.7.108)
where the parameter T ∗ = (T k 2 /gρ) represents the relative importance of surface tension and gravity. It also follows from (2.7.106) that the phase velocity cp has a minimum value at k = km = gρ/T (or T ∗ = 1) with the corresponding minimum value for cp is (cp )m = 1
4T g ρ
14 (2.7.109)
at the wavelength λ = λm = 2π(T /gρ) 2 . The inequality k km is the condition for the waves to be effectively pure gravity waves with negligible surface tension. This is equivalent to large wavelength
2.7 The Euler Equation of Motion and Water Wave Problems
189
Fig. 2.5 The solid curve represents the phase velocity cp for capillary-gravity waves against λ. (From Lighthill 1978.) 1
T 2 λ > λm = ( k2π ) = 2π( ρg ) . The phase velocity cp in (2.7.106) for gravity-capillary m waves in deep water is shown by the solid curve in Figure 2.5 against λ with minimum (cp )m attained at λ = λm . The dotted curve corresponds to (2.7.107) for pure capillary waves (or ripples) in deep water dominated by surface tension for small 1 λ < λm . The dashed curve corresponds to cp = (g/k) 2 for pure gravity waves for large wavelengths λ > λm . The group velocity for gravity-capillary waves can be calculated from the dispersion relation (2.7.102) and is given by g 3T k 2 T k2 2 cg = 1+ tanh kh + 1 + kh sech kh (2.7.110) 2ω gρ gρ
g (1 + 3T ∗ ) tanh kh + (1 + T ∗ )kh sech2 kh . = (2.7.111) 2ω
We next multiply the numerator by kcp and the denominator by ω(= kcp ), then replace ω 2 by (2.7.106) to obtain 1 2kh 1 + 3T ∗ c g = cp + . (2.7.112) 2 1 + T∗ sinh 2kh In the deep water limit, kh → ∞, the second term in (2.7.112) tends to zero, and hence, the group velocity of gravity-capillary waves is 1 + 3T ∗ 1 . (2.7.113) cg = cp 2 1 + T∗ This reduces to the result (2.7.96) for pure gravity waves (T ∗ = 0) in deep water 1 1 g , (2.7.114) cg = cp = 2 2 k
190
2 Nonlinear Model Equations and Variational Principles
and for pure capillary waves in deep water (g → 0 and T ∗ → ∞), the group velocity is 1 3 3 Tk 2 . (2.7.115) cg = cp = 2 2 ρ It follows from the definition of group velocity (2.7.33) that the group and phase velocities are related by a simple formula cg =
dω dcp dcp d = (kcp ) = cp + k = cp − λ . dk dk dk dλ
(2.7.116)
It is clear from (2.7.116) that cg = cp . However, if cp does not depend on the dc wavelength, λ (or wavenumber, k), then cg = cp . If dkp > 0, then cg > cp , and dcp dc if dk < 0, then cg < cp . If cp is minimum for some k,√then dkp = 0, and hence, cg = cp . For shallow water waves, ω 2 = (gh)k 2 or cp = gh, and then cg = cp . In case of gravity-capillary waves in deep water, cp has a minimum for k = kmin = gρ/T , and hence, cg = (cp )m . Figure 2.5 reveals that on the left of the dc minimum, dkp > 0, and hence, result (2.7.116) confirms that cg > cp , whereas on dc the right of the minimum, dkp < 0, and hence, cg < cp . Finally, formula (2.7.116) can also be written in the form −1 −1 dcp ω dcp c g = cp 1 − k = cp 1 − . (2.7.117) dω cp dω This is known as the Rayleigh formula for one-dimensional dispersive waves. The general theory of dispersive waves was developed by Whitham in 1960s that will be discussed in Chapter 7. Example 2.7.5 (Total Energy of Pure Gravity Waves on Water of Constant Depth). We calculate the potential energy and the kinetic energy of pure gravity waves on water constant depth h. The potential energy over a single wavelength λ is given by 1 gρ λ 2 η dx = gρa2 , (2.7.118) V = 2 0 4 where the free surface elevation η given by (2.7.37) is used to obtain the above value V . The kinetic energy T is given by λ n 2 1 φx + φ2z dz dx T = ρ 2 0 −h which can be transformed into λ 1 ∂φ 1 ∂φ T =− ρ φ dx dS = ρ φ 2 ∂n 2 0 ∂z z=0 which is, by using (2.7.30),
2.7 The Euler Equation of Motion and Water Wave Problems
=
1 ρga2 2
λ
sin2 (kx − ωt) dx = 0
1 gρa2 λ. 4
191
(2.7.119)
Hence, the total energy per unit wavelength is E =T +V =
1 gρa2 λ. 2
(2.7.120)
Thus, the total energy is half kinetic and half potential. The horizontal and vertical velocity components of water particles are
agk cosh k(z + h) exp i(ωt − kx) , (2.7.121) u = φx = ω cosh kh
agk sinh k(z + h) v = φz = i exp i(ωt − kx) . (2.7.122) ω cosh kh So, it is possible to determine the actual path of a fluid particle in motion from (2.7.121)–(2.7.122). In terms of particle displacements X and Z of a particle whose mean motion is (x, z), we get X˙ = u and Z˙ = v in which terms of the second order are neglected. So integration gives agk cosh k(z + h) sin(ωt − kx) + X0 , (2.7.123) X= ω2 cosh kh agk sinh k(z + h) cos(ωt − kx) + Z0 , Z= (2.7.124) ω2 cosh kh where X0 and Z0 are constants of integration, and they move the origin of X and Z. Eliminating (ωt − kx) from (2.7.123)–(2.7.124) gives the equation of a particle path as (X − X0 )2 (Z − Z0 )2 a2 + = . (2.7.125) 2 2 cosh k(z + h) sinh k(z + h) sinh2 kh This represents an ellipse with the semi-major axis in the x-direction of magnitude a cosech kh cosh k(z + h) and with semi-minor axis in the z-direction of magnitude a cosech kh cosh k(z +h). Both semi-axes decrease with depth. When X0 = Z0 = 0 and z = −h, X = 0, Z = 0, and particles oscillate along the bottom. But for a real liquid, viscosity would prevent such oscillations. In deep water (kh → ∞), both cosh k(z+h)/ cosh kh and cosh k(z+h)/sinh kh tend to exp(kz); hence, (2.7.123)–(2.7.124) give X − X0 = aekz sin(ωt − kx), Z − Z0 = aekz cos(ωt − kx).
(2.7.126) (2.7.127)
These results show that the paths of the fluid particles are circles of radius aekz . Clearly, the radius decreases exponentially with increasing depth.
192
2 Nonlinear Model Equations and Variational Principles
2.8 The Energy Equation and Energy Flux In dealing with surface gravity waves, it is important and useful to derive an equation that describes the flow of energy in the fluid. Thus, an energy equation is obtained by taking the scalar product of the velocity vector u with the respective terms of the momentum equation (2.7.13), with ∇φ replaced by u so that 1 2 p ∂u + ∇ u + + gz = 0. (2.8.1) ∂t 2 ρ We take the scalar product of u with (2.8.1) and use u · u = u2 to obtain 1 2 p ∂ 1 2 u + u · ∇ u + + gz = 0. ∂t 2 2 ρ
(2.8.2)
Since div u = 0, we can add ( 12 u2 + p/ρ + gz) div u to (2.8.2) and use ∂z/∂t = 0 and the formula for div(au) with any scalar a to derive, multiplying by ρ, ∂ 1 2 1 2 ρu + ρgz + div u ρu + p + gρz = 0. (2.8.3) ∂t 2 2 The terms 12 ρu2 and ρgz represent the kinetic and potential energies, respectively, and equation (2.8.3) describes a balance between the rate of change of energy and energy flux terms, including convection by the velocity and the rate of working of the pressure. In fact, the rate of change of energy per unit volume is described in terms of the divergence of the energy flux F 1 2 (2.8.4) F = u ρu + p + gρz . 2 Equation (2.8.3) gives, writing E = 12 ρu2 + ρgz, ∂E + div F = 0. ∂t
(2.8.5)
This is usually called the law of conservation of energy. In order to see some physical meaning of (2.8.3), we integrate it over some volume V enclosed by a closed surface S. By using the Gauss divergence theorem, we can transform the volume integral over V to a surface integral over S. Consequently, 1 1 p ∂ ρ u2 + gz dV = − div ρu u2 + + gz dV ∂t V 2 2 ρ V 1 2 (2.8.6) ρu + p + ρgz u · n dS. =− S 2 This represents the rate of change of the total energy in a volume V that is equal to the amount of energy flowing out of this volume across the surface S per unit time.
2.9 Exercises
193
For this reason F is called the energy flux density vector. Its magnitude represents the amount of energy passing across a unit surface area normal to the velocity field u per unit time. We may rewrite the right-hand side of (2.8.6) as 1 pu · n dS − ρgzu · n dS. (2.8.7) − u ρu2 · n dS − 2 S S S The first term is the kinetic energy transported across S per unit time by the fluid; the second term is the work done by the pressure forces on the fluid within the surface, and the third term is the work done by the gravitational force acting on the system.
2.9 Exercises 1. Use the Hamilton principle to derive (i) the Newton second law of motion, and (ii) the equation for a simple harmonic oscillator. 2. Derive the equation of motion of an elastic beam of length , line density ρ, cross-sectional moment of inertia I, and modulus of elasticity E which is fixed at each end and performs small transverse oscillations in the horizontal (x, t)plane. 3. Apply the variational principle δ L dx dt = 0, where the Lagrangian L = 12 (u2xx + ut ux ) + u3x , to derive the equation uxt + 6ux uxx + uxxxx = 0. Show that this equation leads to the KdV equation when η = ux . 4. Use the variational principle 1 δ 1 − u2t + u2x 2 dx dt = 0 to derive the Born and Infeld (1934) equation 1 − u2t uxx + 2ux ut uxt − 1 + u2x utt = 0. 5. Show that the variational principle (Whitham 1967a, 1967b) 1 1 1 1 δ ψx ψt + c0 ψx2 + c0 ψx3 + c0 h20 χ2 + 2χx ψx dx dt = 0 2 2 6 12 gives the coupled equations 1 ψxt + c0 (1 + ψx )ψxx + c0 h20 χxx = 0, 6 where c0 and h0 are constants.
ψxx − χ = 0,
194
2 Nonlinear Model Equations and Variational Principles
6. Show that the variational principle (Whitham 1967a, 1967b) 1 δ φt + αβt + u2 + v 2 dx dy dt = 0 2 leads to the equations Dβ − u = 0, Dt
Dα + f v = 0, Dt
ux + vy = 0,
where u = φx + αβx − α, v = φy + αβy − f β, and −p = φt + αβt +
1 2 u + v2 . 2
7. If L = L(ω, k) where ω = −θt and k = θx , show that the variational principle (Whitham 1965a, 1965b; Lighthill 1967) δ L(ω, k) dt dx = 0 gives the Euler–Lagrange equation ∂ ∂ (Lω ) = (Lk ). ∂t ∂x Show also that this equation reduces to a second-order quasi-linear partial differential equation for θ(x, t) Lωω θtt − 2Lωk θxt + Lkk θxx = 0. 8. Derive the Boussinesq equation for water waves utt − c2 uxx − μuxxtt = from the variational principle
1 2 u xx 2
δ
L dx dt = 0,
where L ≡ 12 φ2t − 12 c2 φ2x + 12 μφ2xt − 16 φ3x and φ is the potential for u where u = φx . 9. Show that the Euler equation of the variational principle
δI u(x, y) = δ F (x, y, u, p, q, l, m, n) dx dy = 0 D
is Fu −
∂ ∂ ∂2 ∂2 ∂2 Fp − Fq + Fm + 2 Fn = 0, Fl + 2 ∂x ∂y ∂x ∂x∂y ∂y
where p = ux ,
q = uy ,
l = uxx ,
m = uxy ,
and n = uyy .
2.9 Exercises
195
10. In each of the following cases, apply the variational principle or its simple extension with appropriate boundary conditions to derive the corresponding equations: (a) F = u2xx + u2yy + u2xy , (b) F = 12 [u2t − α2 (u2x + u2y ) − β 2 u2 ], (c) F = 12 (ut ux + αu2x + βu2xx ), (d) F = 12 (u2t + α2 u2xx ), (e) F = p(x)u 2 +
− [r(x) + λs(x)]u2 , where p, q, r, and s are given functions of x, and α and β are constants. 11. Derive the Schrödinger equation from the variational principle 2 2 ψx + ψy2 + ψz2 + (V − E)ψ 2 dx dy dz = 0, δ D 2m d 2 dx (q(x)u )
where h = 2π is the Planck constant, m is the mass of a particle moving under the action of a force field described by the potential V (x, y, z), and E is the total energy of the particle. 12. Derive the Poisson equation ∇2 u = F (x, y) from the variational principle with the functional 2
ux + u2y + 2uF (x, y) dx dy, I(u) = D
where u = u(x, y) is given on the boundary ∂D of D. 13. Prove that the Euler–Lagrange equation for the functional f (x, y, z, u, p, q, r, l, m, n, a, b, c) dx dy dz I= D
is ∂ ∂ ∂ ∂2 ∂2 ∂2 Fp − Fq − Fr + F + F + Fn l m ∂x ∂y ∂z ∂x2 ∂y2 ∂z 2 ∂2 ∂2 ∂2 Fa + Fb + Fc = 0, + ∂x∂y ∂y∂z ∂z∂x
Fu −
where (p, q, r) = (ux , uy , uz ), (l, m, n) = (uxx , uyy , uzz ), and (a, b, c) = (uxy , uyz , uzx ). 14. Derive the equation of motion of a vibrating string of length l under the action of an external force F (x, t) from the variational principle
t2
l
δ t1
0
1 2 ρu − T ∗ u2x 2 t
+ ρuF (x, t) dx dt = 0,
where ρ is the line density and T ∗ is the constant tension of the string.
196
2 Nonlinear Model Equations and Variational Principles
15. The kinetic and potential energies associated with the transverse vibration of a thin elastic plate of constant thickness h are 1 u˙ 2 dx dy, T = ρ 2 D
1 V = μ0 (∇u)2 − 2(1 − σ) uxx uyy − u2xy dx dy, 2 D where ρ is the surface density and μ0 = 2h3 E/3(1 − σ2 ). Use the variational principle t2
(T − V ) + f u dx dy dt = 0 δ t1
D
to derive the equation of motion of the plate ρ¨ u + μ0 ∇4 u = f (x, y, t), where f is the transverse force per unit area acting on the plate. 16. The kinetic and potential energies associated with wave motion in elastic solids are 1 ρ u2t + vt2 + wt2 dx dy dz, T = 2 D 1 λ(ux + vy + wz )2 + 2μ u2x + vy2 + wz2 V = 2 D
+ μ (vx + uy )2 + (wy + vz )2 + (uz + wx )2 dx dy dz. Use the variational principle t2 (T − V ) dx dy dz dt = 0 δ t1
D
to derive the equation of wave motion in an elastic medium (λ + μ) grad div u + μ∇2 u = ρutt , where u = (u, v, w) is the displacement vector. 17. Apply the variational principle (2.5.2) with the Lagrangian L = 12 (ut ux − u2xx − 2u3x ) to derive uxt − 6ux uxx + uxxxx = 0. Show that this equation leads to the KdV equation when ux = η. 18. An inviscid and incompressible fluid flow under the conservative force field F = −∇Ω with a potential Ω = gz, where g is the constant acceleration due to gravity, is governed by the Euler equation (2.7.1). (a) Show that 1 p ∂u + ∇ u · u + + Ω − u × ω = 0, ∂t 2 ρ where ∇ × u = ω is the vorticity vector.
2.9 Exercises
197
(b) Taking the scalar product of the above equation with u, derive the result 1 p ∂ 1 u · u + (u · ∇) u · u + + Ω = 0. ∂t 2 2 ρ (c) Derive the energy equation ∂ 1 1 ρu · u + ρΩ + ∇ · u ρu · u + p + ρΩ = 0. ∂t 2 2 Explain the significance of each term of the energy equation. 19. The three-dimensional Plateau problem is governed by the functional
1 I u(x, y, z) = 1 + p2 + q 2 + r2 2 dx dy dz D
where p = ux , q = uy , and r = uz . Find the Euler–Lagrange equation of this functional. 20. (a) Show that the Euler–Lagrange equation for the functional b F (x, u, u ) dx, I(u) = a
where u = (u1 , u2 , . . . , un ), ui ∈ C 2 [a, b], ui (a) = ai , and ui (b) = bi , i = 1, 2, . . . , n, is a system of n ordinary differential equations Fui −
d Fu = 0, dx i
i = 1, 2, . . . , n.
(b) If F in (a) does not depend explicitly on x, then show that F−
n
ui Fui = const.
i=1
21. Consider a simple pendulum of length with a bob of mass m suspended from a frictionless support. Apply the Hamilton principle to the functional t2 I[θ(t)] = (T − V ) dt, t1
where T = 12 m2 θ˙2 and V = mg( − cos θ) to derive the equation of the simple pendulum g θ¨ + ω2 sin θ = 0, ω 2 = . 22. Derive the Euler–Lagrange equation for the functional b I(y(x)) = F (x, y, y ) dx, a
where
198
2 Nonlinear Model Equations and Variational Principles
(a) F (x, y, y ) = u(x, y) 1 + y 2 , 2 1 ) 2 with y(a) = y1 , y(b) = y2 < y1 (Brachis(b) F (x, y, y ) = √12g ( 1+y y1 −y tochrone problem). 23. The Fermat principle in optics states that light travels from one point A(x1 , y1 ) to another point B(x2 , y2 ) in an optically homogeneous medium along a path in a minimum (least) time. The time taken for the light beam to travel from A to B is t2 x2 dt 1 x2 ds dx = I y(x) = dt = 1 + y 2 dx, ds dx c x1 t1 x1 is the constant velocity of light. where c = ds dt Apply the variational principle 1 1 + y 2 dx = 0 δI = δ c to derive the Snell law of refraction of light, n sin φ = const., where n = 1c is the refractive index of the medium and φ is the angle made by the tangent to the minimum path with the vertical y-axis. 24. (a) Derive the principle of least action for a conservative system t2 2T dt = 0, δ t1
where the time integral of 2T is called the action of the system. (b) Explain the significance of this principle. 25. Show that the Euler–Lagrange equation of the variational principle δI = δ
b
F x, y, y , y , . . . , y (n) dx = 0
a
is an ordinary differential equation of order 2n Fy −
d d2 dn Fy + 2 Fy − · · · + (−1)n n Fy(n) = 0. dx dx dx
26. The electrostatic potential φ(x, y, z) is defined in terms of the electrostatic field E so that E = −∇φ in a domain D of volume V , where φ is specified on ∂D. Show that φ that minimizes the electric energy functional e0 e0 I[φ] = E 2 dv = (∇φ)2 dv 2 2 V V satisfies the Laplace equation. 27. Use the Lagrangian L = T − V and the Lagrange equation to derive the Newton laws of motion of a particle of mass m moving under a force, F = −∇V , in
2.9 Exercises
199
(a) one dimension, (b) a two dimensional plane in Cartesian coordinates, (c) a two dimensional plane in polar coordinates. 28. Seek a traveling wave solution rn = A cos θ = A cos(ωt − kn) of the linearized Toda lattice equation m¨ rn = (ab)(rn+1 − 2rn + rn−1 ), where rn = (yn+1 − yn ) and yn is the displacement of the nth mass. Show that the dispersion relation is 4ab 2 k 2 ω = sin . m 2 29. Consider the Ablowitz and Ladik (AL) equation for the lattice system (1976a, 1976b) γ dφn 2 + (φn+1 + φn−1 ) 1 + |φn | = 0. i dt 2 (a) Using φn = e2it ψn , show that the solution of the AL equation reduces to that of the NLS equation iψt + ψxx + γ|ψ|2 ψ = 0, as the ratio of anharmoncity to dispersion (γ) tends to zero. (b) Show that the solution of the above AL equation is
φn (t) = Acn β(n − vt); k exp −i(ωt + αn + φ0 ) , selecting the units of φn so that A=
ksn(β; k) , dn(β; k)
ω=−
γ 2
= 1, the parameters A, ω and v are given by
2cn(β; k) cos α , dn2 (β; k)
where 0 < β < ∞, −π ≤ α ≤ π, and 0 < k < 1.
v=−
2sn(β; k) sin α , β dn(β; k)
3 First-Order, Quasi-linear Equations and Method of Characteristics
As long as a branch of knowledge offers an abundance of problems, it is full of vitality. David Hilbert The advance of analysis is, at this moment, stagnant along the entire front of nonlinear problems. That this phenomenon is not of a transient nature but that we are up against an important conceptual difficulty . . . yet no decisive progress has been made against them . . . which could be rated as important by the criteria that are applied in other, more successful (linear!) parts of mathematical physics. It is important to avoid a misunderstanding at this point. One may be tempted to qualify these (shock wave and turbulence) problems as problems in physics, rather than in applied mathematics, or even pure mathematics. We wish to emphasize that it is our conviction that such an interpretation is wholly erroneous. John Von Neumann
3.1 Introduction Many problems in mathematical, physical, and engineering sciences deal with the formulation and the solution of first-order partial differential equations. From a mathematical point of view, first-order equations have the advantage of providing a conceptual basis that can be utilized for second-, third-, and higher-order equations. This chapter is concerned with first-order, quasi-linear and linear partial differential equations and their solutions by using the Lagrange method of characteristics and its generalizations.
L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_3,
202
3 First-Order, Quasi-linear Equations and Method of Characteristics
3.2 The Classification of First-Order Equations The most general, first-order, partial differential equation in two independent variables x and y is of the form F (x, y, u, ux , uy ) = 0,
(x, y) ∈ D ⊂ R2 ,
(3.2.1)
where F is a given function of its arguments, and u = u(x, y) is an unknown function of the independent variables x and y which lie in some given domain D in R2 , and uy = ∂u . Equation (3.2.1) is often written in terms of standard notation ux = ∂u ∂x ∂y p = ux and q = uy so that (3.2.1) takes the form F (x, y, u, p, q) = 0.
(3.2.2)
Similarly, the most general, first-order, partial differential equation in three independent variables x, y, z can be written as F (x, y, z, u, ux , uy , uz ) = 0.
(3.2.3)
Equation (3.2.1) or (3.2.2) is called a quasi-linear partial differential equation if it is linear in first-partial derivatives of the unknown function u(x, y). So, the most general quasi-linear equation must be of the form a(x, y, u)ux + b(x, y, u)uy = c(x, y, u), where its coefficients a, b, and c are functions of x, y, and u. The following are examples of quasi-linear equations: x y 2 + u ux − y x2 + u uy = x2 − y 2 u, 2
uux + ut + nu = 0, 2 y − u2 ux − xyuy = xu.
(3.2.4)
(3.2.5) (3.2.6) (3.2.7)
Equation (3.2.4) is called a semilinear partial differential equation if its coefficients a and b are independent of u, and hence, the semilinear equation can be expressed in the form a(x, y)ux + b(x, y)uy = c(x, y, u).
(3.2.8)
Examples of semilinear equations are xux + yuy = u2 + x2 , (x + 1)2 ux + (y − 1)2 uy = (x + y)u2 , ut + aux + u2 = 0,
(3.2.9) (3.2.10) (3.2.11)
where a is a constant. Equation (3.2.1) is said to be linear if F is linear in each of the variables u, ux , and uy , and the coefficients of these variables are functions only of the independent
3.3 The Construction of a First-Order Equation
203
variables x and y. The most general, first-order, linear partial differential equation has the form (3.2.12) a(x, y)ux + b(x, y)uy + c(x, y)u = d(x, y), where the coefficients a, b, and c, in general, are functions of x and y, and d(x, y) is a given function. Unless stated otherwise, these functions are assumed to be continuously differentiable. Equations of the form (3.2.12) are called homogeneous if d(x, y) ≡ 0 or inhomogeneous if d(x, y) = 0. Obviously, linear equations are a special kind of the quasi-linear equation (3.2.4) if a, b are independent of u and c is a linear function in u. Similarly, semilinear equation (3.2.8) reduces to a linear equation if c is linear in u. Examples of linear equations are xux + yuy − nu = 0, nux + (x + y)uy − u = ex , yux + xuy = xy, (y − z)ux + (z − x)uy + (x − y)uz = 0.
(3.2.13) (3.2.14) (3.2.15) (3.2.16)
An equation which is not linear is often called a nonlinear equation. So, the first-order equations are often classified as linear and nonlinear.
3.3 The Construction of a First-Order Equation We consider a system of geometrical surfaces described by the equation f (x, y, z, a, b) = 0,
(3.3.1)
where a and b are arbitrary parameters. We differentiate (3.3.1) with respect to x and y to obtain fy + qfz = 0, (3.3.2) fx + pfz = 0, ∂z ∂z where p = ∂x and q = ∂y . The set of three equations (3.3.1) and (3.3.2) involves two arbitrary parameters a and b. In general, these two parameters can be eliminated from this set to obtain a first-order equation of the form
F (x, y, z, p, q) = 0.
(3.3.3)
Thus the system of surfaces (3.3.1) gives rise to a first-order partial differential equation (3.3.3). In other words, an equation of the form (3.3.1) containing two arbitrary parameters is called a complete solution or a complete integral of equation (3.3.3). Its role is somewhat similar to that of a general solution for the case of an ordinary differential equation. On the other hand, any relationship of the form f (φ, ψ) = 0,
(3.3.4)
204
3 First-Order, Quasi-linear Equations and Method of Characteristics
which involves an arbitrary function f of two known functions φ = φ(x, y, z) and ψ = ψ(x, y, z) and provides a solution of a first-order partial differential equation is called a general solution or general integral of this equation. Clearly, the general solution of a first-order partial differential equation depends on an arbitrary function. This is in striking contrast to the situation for ordinary differential equations where the general solution of a first-order ordinary differential equation depends on one arbitrary constant. The general solution of a partial differential equation can be obtained from its complete integral. We obtain the general solution of (3.3.3) from its complete integral (3.3.1) as follows. First, we prescribe the second parameter b as an arbitrary function of the first parameter a in the complete solution (3.3.1) of (3.3.3), that is, b = b(a). We then consider the envelope of the one-parameter family of solutions so defined. This envelope is represented by the two simultaneous equations f x, y, z, a, b(a) = 0, (3.3.5) (3.3.6) fa x, y, z, a, b(a) + fb x, y, z, b(a) b (a) = 0, where the second equation (3.3.6) is obtained from the first equation (3.3.5) by partial differentiation with respect to a. In principle, equation (3.3.5) can be solved for a = a(x, y, z) as a function of x, y, and z. We substitute this result back in (3.3.5) to obtain f x, y, z, a(x, y, z), b a(x, y, z) = 0, (3.3.7) where b is an arbitrary function. Indeed, the two equations (3.3.5) and (3.3.6) together define the general solution of (3.3.3). When a definite b(a) is prescribed, we obtain a particular solution from the general solution. Since the general solution depends on an arbitrary function, there are infinitely many solutions. In practice, only one solution satisfying prescribed conditions is required for a physical problem. Such a solution may be called a particular solution. In addition to the general and particular solutions of (3.3.3), if the envelope of the two-parameter system (3.3.1) of surfaces exists, it also represents a solution of the given equation (3.3.3); the envelope is called the singular solution of equation (3.3.3). The singular solution can easily be constructed from the complete solution (3.3.1) representing a two-parameter family of surfaces. The envelope of this family is given by the system of three equations f (x, y, z, a, b) = 0,
fa (x, y, z, a, b) = 0,
fb (x, y, z, a, b) = 0.
(3.3.8)
In general, it is possible to eliminate a and b from (3.3.8) to obtain the equation of the envelope which gives the singular solution. It may be pointed out that the singular solution cannot be obtained from the general solution. Its nature is similar to that of the singular solution of a first-order ordinary differential equation. Finally, it is important to note that solutions of a partial differential equation are expected to be represented by smooth functions. A function is called smooth if all of its derivatives exist and are continuous. However, in general, solutions are not always smooth. A solution which is not everywhere differentiable is called a weak solution.
3.3 The Construction of a First-Order Equation
205
The most common weak solution is the one that has discontinuities in its first partial derivatives across a curve, so that the solution can be represented by shock waves as surfaces of discontinuity. In the case of a first-order partial differential equation, there ∂z ∂z and q = ∂y are are discontinuous solutions where z itself and not merely p = ∂x discontinuous. In fact, this kind of discontinuity is usually known as a shock wave. An important feature of quasi-linear and nonlinear partial differential equations is that their solutions may develop discontinuities as they move away from the initial state. We close this section by considering some examples. Example 3.3.1. Show that a family of spheres x2 + y 2 + (z − c)2 = r2 ,
(3.3.9)
satisfies the first-order linear partial differential equation yp − xq = 0.
(3.3.10)
Differentiating the equation (3.3.9) with respect to x and y gives x + p(z − c) = 0 and
y + q(z − c) = 0.
Eliminating the arbitrary constant c from these equations, we obtain the first-order, partial differential equation yp − xq = 0. Example 3.3.2. Show that the family of spheres (x − a)2 + (y − b)2 + z 2 = r2 satisfies the first-order, nonlinear, partial differential equation z 2 p2 + q 2 + 1 = r 2 .
(3.3.11)
(3.3.12)
We differentiate the equation of the family of spheres with respect to x and y to obtain (x − a) + zp = 0, (y − b) + z q = 0. Eliminating the two arbitrary constants a and b, we find the nonlinear partial differential equation z 2 p2 + q 2 + 1 = r 2 . All surfaces of revolution with the z-axis as the axis of symmetry satisfy the equation (3.3.13) z = f x2 + y 2 , where f is an arbitrary function. Writing u = x2 + y 2 and differentiating (3.3.13) with respect to x and y, respectively, we obtain p = 2xf (u),
q = 2yf (u).
Eliminating the arbitrary function f (u) from these results, we find the equation yp − xq = 0.
206
3 First-Order, Quasi-linear Equations and Method of Characteristics
Theorem 3.3.1. If φ = φ(x, y, z) and ψ = ψ(x, y, z) are two given functions of x, y, and z and if f (φ, ψ) = 0, where f is an arbitrary function of φ and ψ, then z = z(x, y) satisfies a first-order, partial differential equation p where
∂(φ, ψ) ∂(φ, ψ) ∂(φ, ψ) +q = , ∂(y, z) ∂(z, x) ∂(x, y) ∂(φ, ψ) φx = ψx ∂(x, y)
φy . ψy
(3.3.14)
(3.3.15)
Proof. We differentiate f (φ, ψ) = 0 with respect to x and y, respectively, to obtain the following equations: ∂f ∂ψ ∂φ ∂ψ ∂f ∂φ + = 0, (3.3.16) +p +p ∂φ ∂x ∂z ∂ψ ∂x ∂z ∂f ∂ψ ∂f ∂φ ∂φ ∂ψ + = 0. (3.3.17) +q +q ∂φ ∂y ∂z ∂ψ ∂y ∂z ∂f ∂f Nontrivial solutions for ∂φ and ∂ψ can be found if the determinant of the coefficients of these equations vanishes, that is, φx + pφz ψx + pψz (3.3.18) φy + qφz ψy + qψz = 0.
Expanding this determinant gives the first-order, quasi-linear equation (3.3.14).
3.4 The Geometrical Interpretation of a First-Order Equation To investigate the geometrical content of a first-order, partial differential equation, we begin with a general, quasi-linear equation a(x, y, u)ux + b(x, y, u)uy − c(x, y, u) = 0.
(3.4.1)
We assume that the possible solution of (3.4.1) in the form u = u(x, y) or in an implicit form f (x, y, u) ≡ u(x, y) − u = 0 (3.4.2) represents a possible solution surface in the (x, y, u)-space. This is often called an integral surface of the equation (3.4.1). At any point (x, y, u) on the solution surface, the gradient vector ∇f = (fx , fy , fu ) = (ux , uy , −1) is normal to the solution surface. Clearly, equation (3.4.1) can be written as the dot product of two vectors aux + buy − c = (a, b, c) · (ux , uy , −1) = 0.
(3.4.3)
This clearly shows that the vector (a, b, c) must be a tangent vector of the integral surface (3.4.2) at the point (x, y, u), and hence, it determines a direction field called
3.4 The Geometrical Interpretation of a First-Order Equation
207
Fig. 3.1 Tangent and normal vector fields of solution surface at a point (x, y, u).
the characteristic direction or Monge axis. This direction is of fundamental importance in determining a solution of equation (3.4.1). To summarize, we have shown that f (x, y, u) = u(x, y) − u = 0, as a surface in the (x, y, u)-space, is a solution of (3.4.1) if and only if the direction vector field (a, b, c) lies in the tangent planes of the integral surface f (x, y, u) = 0 at each point (x, y, u), where ∇f = 0, as shown in Figure 3.1. A curve in the (x, y, u)-space whose tangent at every point coincides with the characteristic direction field (a, b, c) is called a characteristic curve. If the parametric equations of this characteristic curve are x = x(t),
y = y(t),
u = u(t),
(3.4.4)
dy du then the tangent vector to this curve is ( dx dt , dt , dt ) which must be equal to (a, b, c). Therefore, the system of ordinary differential equations of the characteristic curve is given by
dx = a(x, y, u), dt
dy = b(x, y, u), dt
du = c(x, y, u). dt
(3.4.5)
These are called the characteristic equations of the quasi-linear equation (3.4.1). In fact, there are only two independent ordinary differential equations in the system (3.4.5); therefore, its solutions form a two-parameter family of curves in the (x, y, u)-space. The projection on u = 0 of a characteristic curve on the (x, t)-plane is called a characteristic base curve, or simply a characteristic. Equivalently, the characteristic equations (3.4.5) in the nonparametric form are dy du dx = = . a b c
(3.4.6)
The typical problem of solving equation (3.4.1) with a prescribed u on a given plane curve C is equivalent to finding an integral surface in the (x, y, u)-space, satisfying equation (3.4.1) and containing the three-dimensional space curve Γ defined by the values of u on C, which is the projection of Γ on u = 0.
208
3 First-Order, Quasi-linear Equations and Method of Characteristics
Remark 1. The above geometrical interpretation can be generalized for higher-order partial differential equations. However, it is not easy to visualize geometrical arguments that have been described for the case of three space dimensions. Remark 2. The geometrical interpretation is more complicated for the case of nonlinear partial differential equations because the normals to possible solution surfaces through a point do not lie in a plane. The tangent planes no longer intersect along one straight line, but instead, they envelope along a curved surface known as the Monge cone. Any further discussion is beyond the scope of this book. We conclude this section by adding an important observation regarding the nature of the characteristics in the (x, t)-plane. For a quasi-linear equation, characteristics are determined by the first two equations in (3.4.5) with their slopes b(x, y, u) dy = . dx a(x, y, u)
(3.4.7)
If (3.4.1) is a linear equation, then a and b are independent of u, and the characteristics of (3.4.1) are plane curves with slopes dy b(x, y) = . dx a(x, y)
(3.4.8)
By integrating this equation, we can determine the characteristics which represent a one-parameter family of curves in the (x, t)-plane. However, if a and b are constant, the characteristics of equation (3.4.1) are straight lines.
3.5 The Method of Characteristics and General Solutions We can use the geometrical interpretation of the first-order, partial differential equation and the properties of characteristic curves to develop a method for finding the general solution of quasi-linear equations. This is usually referred to as the method of characteristics due to Lagrange. This method of solution of quasi-linear equations can be described by the following result. Theorem 3.5.1. The general solution of a first-order, quasi-linear partial differential equation (3.5.1) a(x, y, u)ux + b(x, y, u)uy = c(x, y, u) is f (φ, ψ) = 0,
(3.5.2)
where f is an arbitrary function of φ(x, y, u) and ψ(x, y, u) and φ = const. = c1 and ψ = const. = c2 are solution curves of the characteristic equations dx dy du = = . a b c
(3.5.3)
The solution curves defined by φ(x, y, u) = c1 and ψ(x, y, u) = c2 are called the families of characteristic curves of equation (3.5.1).
3.5 The Method of Characteristics and General Solutions
209
Proof. Since φ(x, y, u) = c1 and ψ(x, y, u) = c2 satisfy equations (3.5.3), these equations must be compatible with the equation dφ = φx dx + φy dy + φu du = 0.
(3.5.4)
This is equivalent to the equation (3.5.5)
a φx + b φy + c φu = 0. Similarly, equation (3.5.3) is also compatible with
(3.5.6)
a ψx + b ψy + c ψu = 0. We now solve (3.5.5), (3.5.6) for a, b, and c to obtain a ∂(φ,ψ) ∂(y,u)
=
b ∂(φ,ψ) ∂(u,x)
=
c ∂(φ,ψ) ∂(x,y)
.
(3.5.7)
It has been shown earlier that f (φ, ψ) = 0 satisfies an equation similar to (3.3.14), that is, ∂(φ, ψ) ∂(φ, ψ) ∂(φ, ψ) p +q = . (3.5.8) ∂(y, u) ∂(u, x) ∂(x, y) Substituting (3.5.7) into (3.5.8), we find that f (φ, ψ) = 0 is a solution of (3.5.1). This completes the proof. Note that an analytical method has been used to prove Theorem 3.5.1. Alternatively, a geometrical argument can be used to prove this theorem. The geometrical method of proof is left to the reader as an exercise. Many problems in applied mathematics, science, and engineering involve partial differential equations. We seldom try to find or discuss the properties of a solution to these equations in its most general form. In most cases of interest, we deal with those solutions of partial differential equations which satisfy certain supplementary conditions. In the case of a first-order partial differential equation, we determine the specific solution by formulating an initial-value problem or a Cauchy problem. Theorem 3.5.2 (The Cauchy Problem for a First-Order Partial Differential Equation). Suppose that C is a given curve in the (x, y)-plane with its parametric equations y = y0 (t), (3.5.9) x = x0 (t), where t belongs to an interval I ⊂ R, and the derivatives x0 (t) and y0 (t) are piece2 wise continuous functions such that x2 0 + y0 = 0. Also, suppose that u = u0 (t) is a given function on the curve C. Then, there exists a solution u = u(x, y) of the equation F (x, y, u, ux , uy ) = 0 (3.5.10) in a domain D of R2 containing the curve C for all t ∈ I, and the solution u(x, y) satisfies the given initial data, that is,
210
3 First-Order, Quasi-linear Equations and Method of Characteristics
u x0 (t), y0 (t) = u0 (t)
(3.5.11)
for all values of t ∈ I. In short, the Cauchy problem is to determine a solution of equation (3.5.10) in a neighborhood of C, such that the solution u = u(x, y) takes a prescribed value u0 (t) on C. The curve C is called the initial curve of the problem, and u0 (t) is called the initial data. Equation (3.5.11) is called the initial condition of the problem. The solution of the Cauchy problem also deals with such questions as the conditions on the functions F , x0 (t), y0 (t), and u0 (t) under which a solution exists and is unique. We next discuss a method for solving a Cauchy problem for the first-order, quasilinear equation (3.5.1). We first observe that geometrically x = x0 (t), y = y0 (t), and u = u0 (t) represent an initial curve Γ in the (x, y, u)-space. The curve C, on which the Cauchy data is prescribed, is the projection of Γ on the (x, y)-plane. We now present a precise formulation of a Cauchy problem for the first-order, quasilinear equation (3.5.1). Theorem 3.5.3 (The Cauchy Problem for a Quasi-linear Equation). Suppose that x0 (t), y0 (t), and u0 (t) are continuously differentiable functions of t in a closed interval, 0 ≤ t ≤ 1, and that a, b, and c are functions of x, y, and u with continuous first-order partial derivatives with respect to their arguments in some domain D of the (x, y, u)-space containing the initial curve Γ : x = x0 (t),
y = y0 (t),
u = u0 (t),
where 0 ≤ t ≤ 1, and satisfying the condition y0 (t)a x0 (t), y0 (t), u0 (t) − x0 (t)b x0 (t), y0 (t), u0 (t) = 0.
(3.5.12)
(3.5.13)
Then there exists a unique solution u = u(x, y) of the quasi-linear equation (3.5.1) in the neighborhood of C : x = x0 (t), y = y0 (t), and the solution satisfies the initial condition (3.5.14) u0 (t) = u x0 (t), y0 (t) for 0 ≤ t ≤ 1. Note: Condition (3.5.13) excludes the possibility of C being a characteristic. Example 3.5.1. Find the general solution of the first-order linear partial differential equation (3.5.15) xux + yuy = u. The characteristic curves of this equation are the solutions of the characteristic equations dx dy du = = . (3.5.16) x y u This system of equations gives the integral surfaces φ=
y = C1 x
and
ψ=
u = C2 , x
3.5 The Method of Characteristics and General Solutions
211
where C1 and C2 are arbitrary constants. Thus, the general solution of (3.5.15) is y u = 0, (3.5.17) , f x x where f is an arbitrary function. This general solution can also be written as y u(x, y) = xg , (3.5.18) x where g is an arbitrary function. Example 3.5.2. Obtain the general solution of the linear Euler equation xux + yuy = nu.
(3.5.19)
The integral surfaces are the solutions of the characteristic equations dx dy du = = . x y nu
(3.5.20)
From these equations, we get y = C1 , x
u = C2 , xn
where C1 and C2 are arbitrary constants. Hence, the general solution of (3.5.19) is y u = 0. (3.5.21) f , x xn This can also be written as
u y , = g xn x
or equivalently, u(x, y) = xn g
y . x
(3.5.22)
This shows that the solution u(x, y) is a homogeneous function of x and y of degree n. Example 3.5.3. Find the general solution of the linear equation x2 ux + y 2 uy = (x + y)u.
(3.5.23)
The characteristic equations associated with (3.5.23) are dx dy du . = 2 = x2 y (x + y)u From the first two of these equations, we find
(3.5.24)
212
3 First-Order, Quasi-linear Equations and Method of Characteristics
x−1 − y −1 = C1 ,
(3.5.25)
where C1 is an arbitrary constant. It follows from (3.5.24) that du dx − dy = , 2 2 x −y (x + y)u or equivalently,
du d(x − y) = . x−y u
This gives
x−y = C2 , u where C2 is a constant. Furthermore, (3.5.25) and (3.5.26) also give xy = C3 , u where C3 is a constant. Thus, the general solution (3.5.23) is given by xy x − y = 0, , f u u
(3.5.26)
(3.5.27)
(3.5.28)
where f is an arbitrary function. This general solution representing the integral surface can also be written as x−y , (3.5.29) u(x, y) = xyg u where g is an arbitrary function, or equivalently, x−y , u(x, y) = xyh xy
(3.5.30)
where h is an arbitrary function. Example 3.5.4. Show that the general solution of the linear equation
is
(y − z)ux + (z − x)uy + (x − y)uz = 0
(3.5.31)
u(x, y, z) = f x + y + z, x2 + y 2 + z 2 ,
(3.5.32)
where f is an arbitrary function. The characteristic curves satisfy the characteristic equations dy dz du dx = = = , y−z z−x x−y 0
(3.5.33)
3.5 The Method of Characteristics and General Solutions
213
or equivalently, du = 0,
dx + dy + dz = 0,
x dx + y dy + z dz = 0.
Integration of these equations gives u = C1 ,
x + y + z = C2 ,
x2 + y 2 + z 2 = C3 ,
and
where C1 , C2 and C3 are arbitrary constants. Thus, the general solution can be written in terms of an arbitrary function f in the form u(x, y, z) = f x + y + z, x2 + y 2 + z 2 . We next verify that this is a general solution by introducing three independent variables ξ, η, ζ defined in terms of x, y, and z as η = x2 + y 2 + z 2 ,
ξ = x + y + z,
and ζ = y + z,
(3.5.34)
where ζ is an arbitrary combination of y and z. Clearly, the general solution becomes u = f (ξ, η), and hence, uζ = ux
∂x ∂y ∂z + uy + uz . ∂ζ ∂ζ ∂ζ
(3.5.35)
It follows from (3.5.34) that ∂x ∂y ∂z + + , 0= ∂ζ ∂ζ ∂ζ
∂x ∂y ∂z , 0=2 x +y +z ∂ζ ∂ζ ∂ζ
It follows from the first and the third results that x=y
∂z ∂y +z , ∂ζ ∂ζ
y=y
∂x ∂ζ
∂y ∂z +y , ∂ζ ∂ζ
∂y ∂z + = 1. ∂ζ ∂ζ
= −1 and, therefore, z=z
∂y ∂z +z . ∂ζ ∂ζ
Clearly, it follows by subtracting that x − y = (z − y) Using the values for (z − y)
∂x ∂z ∂ζ , ∂ζ ,
and
∂z , ∂ζ
∂y ∂ζ
x − z = (y − z)
∂y . ∂ζ
in (3.5.35), we obtain
∂u ∂u ∂u ∂u = (y − z) + (z − x) + (x − y) . ∂ζ ∂x ∂y ∂z
(3.5.36)
If u = u(ξ, η) satisfies (3.5.31), then ∂u ∂ζ = 0, and hence, (3.5.36) reduces to (3.5.31). This shows that the general solution (3.5.32) satisfies equation (3.5.31).
214
3 First-Order, Quasi-linear Equations and Method of Characteristics
Example 3.5.5. Find the solution of the equation u(x + y)ux + u(x − y)uy = x2 + y 2 ,
(3.5.37)
with the Cauchy data u = 0 on y = 2x. The characteristic equations are dx dy du x dx − y dy − u du y dx + x dy − u du = = 2 = . = 2 u(x + y) u(x − y) x +y 0 0 Consequently,
1 d xy − u2 =0 2
and
d
1 2 x − y 2 − u2 = 0. 2
(3.5.38)
These give two integrals u2 − x2 + y 2 = C1
and 2xy − u2 = C2 ,
(3.5.39ab)
where C1 and C2 are constants. Hence, the general solution is f x2 − y 2 − u2 , 2xy − u2 = 0, where f is an arbitrary function. Using the Cauchy data in (3.5.39ab), we obtain 4C1 = 3C2 . Therefore, 4 u2 − x2 + y 2 = 3 2xy − u2 . Thus, the solution of equation (3.5.37) is given by 7u2 = 6xy + 4 x2 − y 2 .
(3.5.40)
Example 3.5.6. Obtain the solution of the linear equation ux − uy = 1,
(3.5.41)
with the Cauchy data u(x, 0) = x2 . The characteristic equations are dx dy du = = . 1 −1 1 Obviously, dy = −1 dx
and
du = 1. dx
Clearly, x + y = const. = C1 Thus, the general solution is given by
and
u − x = const. = C2 .
(3.5.42)
3.5 The Method of Characteristics and General Solutions
215
Fig. 3.2 Characteristics of equation (3.5.41).
u − x = f (x + y),
(3.5.43)
where f is an arbitrary function. We now use the Cauchy data to find f (x) = x2 − x, and hence, the solution is u(x, y) = (x + y)2 − y.
(3.5.44)
The characteristics x+y = C1 are drawn in Figure 3.2. The value of u must be given at one point on each characteristic which intersects the line y = 0 only at one point, as shown in Figure 3.2. Example 3.5.7. Obtain the solution of the equation (y − u)ux + (u − x)uy = x − y,
(3.5.45)
with the condition u = 0 on xy = 1. The characteristic equations for equation (3.5.45) are dy du dx = = . y−u u−x x−y
(3.5.46)
The parametric forms of these equations are dx = y − u, dt
dy = u − x, dt
du = x − y. dt
These lead to the following equations: x˙ + y˙ + u˙ = 0 and
xx˙ + y y˙ + uu˙ = 0,
(3.5.47ab)
where the dot denotes the derivative with respect to t. Integrating (3.5.47ab), we obtain x + y + u = const. = C1
and x2 + y 2 + u2 = const. = C2 .
These equations represent circles.
(3.5.48ab)
216
3 First-Order, Quasi-linear Equations and Method of Characteristics
Using the Cauchy data, we find that C12 = (x + y)2 = x2 + y 2 + 2xy = C2 + 2. Thus, the integral surface is described by (x + y + u)2 = x2 + y 2 + u2 + 2. Hence, the solution is given by u(x, y) =
1 − xy . x+y
(3.5.49)
Example 3.5.8. Solve the linear equation yux + xuy = u,
(3.5.50)
with the Cauchy data u(x, 0) = x3
and
u(0, y) = y 3 .
(3.5.51ab)
The characteristic equations are dy du dx = = , y x u or equivalently,
dx − dy dx + dy du = = . u y−x y+x
Solving these equations, we obtain u=
C1 = C2 (x + y), x−y
or u = C2 (x + y),
x2 − y 2 =
C1 = const. = C. C2
So the characteristics are rectangular hyperbolas for C > 0 or C < 0. Thus, the general solution is given by u 2 2 ,x − y = 0. f x+y Or equivalently,
u(x, y) = (x + y)g x2 − y 2 . 2
2
Using the Cauchy data, we find that g(x ) = x , that is, g(x) = x.
(3.5.52)
3.5 The Method of Characteristics and General Solutions
Consequently, the solution becomes u(x, y) = (x + y) x2 − y 2 Similarly,
u(x, y) = (x + y) y 2 − x2
217
on x2 − y2 = C > 0. on y 2 − x2 = C > 0.
It follows from these results that u → 0 in all regions, as x → ±y (or y → ±x), and hence, u is continuous across y = ±x which represent asymptotes of the rectangular hyperbolas x2 − y 2 = C. However, ux and uy are not continuous, as y → ±x. For x2 − y 2 = C > 0, ux = 3x2 + 2xy − y 2 = (x + y)(3x − y) → 0, as y → −x, uy = −3y2 − 2xy + x2 = (x + y)(x − 3y) → 0, as y → −x. Hence, both ux and uy are continuous as y → −x. On the other hand, ux → 4x2 ,
uy → −4x2
as y → x.
This implies that ux and uy are discontinuous across y = x. Combining all these results, we conclude that u(x, y) is continuous everywhere in the (x, t)-plane, and ux , uy are continuous everywhere in the (x, t)-plane except on the line y = x. Hence, the partial derivatives ux , uy are discontinuous on y = x. Thus, the development of discontinuities across characteristics is a significant feature of the solutions of partial differential equations. Example 3.5.9. Determine the integral surfaces of the equation x y 2 + u ux − y x2 + u uy = x2 − y 2 u,
(3.5.53)
with the data x + y = 0,
u = 1.
The characteristic equations are dx dy du = = 2 , x(y 2 + u) −y(x2 + u) (x − y 2 )u
(3.5.54)
or equivalently, dx x
(y 2 + u)
=
dy y 2 −(x +
u)
=
du u
(x2 − y 2 )
Consequently, log(xyu) = log C1 , or xyu = C1 .
=
dx x
+
dy y
0
+
du u
.
218
3 First-Order, Quasi-linear Equations and Method of Characteristics
From (3.5.54), we obtain y dy du x dx + y dy − du x dx = = 2 = , x2 (y 2 + u) −y 2 (x2 + u) (x − y 2 )u 0 whence we find that x2 + y 2 − 2u = C2 . Using the given data, we obtain C1 = −x2
and
C2 = 2x2 − 2,
so that C2 = −2(C1 + 1). Thus the integral surface is given by x2 + y 2 − 2u = −2 − 2xyu, or 2xyu + x2 + y 2 − 2u + 2 = 0.
(3.5.55)
Example 3.5.10. Obtain the solution of the equation xux + yuy = x exp(−u)
(3.5.56)
with the data u = 0 on y = x2 . The characteristic equations are dx dy du = = , x y x exp(−u)
(3.5.57)
or
y = C1 . x We also obtain from (3.5.57) that dx = eu du which can be integrated to find eu = x + C2 . Thus, the general solution is given by y u f e − x, =0 x or equivalently,
y e =x+g . (3.5.58) x Applying the Cauchy data, we obtain g(x) = 1 − x. Thus, the solution of (3.5.56) is given by y eu = x + 1 − , x or y . (3.5.59) u = log x + 1 − x u
3.5 The Method of Characteristics and General Solutions
219
Example 3.5.11. Solve the initial-value problem ut + u ux = x,
u(x, 0) = f (x),
(3.5.60)
where (a) f (x) = 1 and (b) f (x) = x. The characteristic equations are dx du d(x + u) dt = = = . 1 u x x+u
(3.5.61)
Integration gives t = log(x + u) − log C1 , or
(u + x)e−t = C1 .
Similarly, we get u2 − x2 = C2 . For case (a), we obtain 1 + x = C1 Thus, or equivalently,
and 1 − x2 = C2 ,
and hence,
C2 = 2C1 − C12 .
2 u − x2 = 2(u + x)e−t − (u + x)2 e−2t , u − x = 2e−t − (u + x)e−2t .
A simple manipulation gives the solution u(x, t) = x tanh t + sech t.
(3.5.62)
Case (b) is left to the reader as an exercise. Example 3.5.12. Find the integral surface of the equation (3.5.63)
u ux + uy = 1,
so that the surface passes through an initial curve represented parametrically by x = x0 (s),
y = y0 (s),
u = u0 (s),
where s is a parameter. The characteristic equations for the given equations are dy du dx = = , u 1 1
(3.5.64)
220
3 First-Order, Quasi-linear Equations and Method of Characteristics
which are, in the parametric form, dy = 1, dτ
dx = u, dτ
du = 1, dτ
(3.5.65)
where τ is a parameter. Thus the solutions of this parametric system in general depend on two parameters s and τ . We solve this system (3.5.65) with the initial data x(s, 0) = x0 (s),
y(s, 0) = y0 (s),
u(s, 0) = u0 (s).
The solutions of (3.5.65) with the given initial data are ⎫ 2 x(s, τ ) = τ2 + τ u0 (s) + x0 (s), ⎪ ⎬ y(s, τ ) = τ + y0 (s), ⎪ ⎭ u(s, τ ) = τ + u0 (s).
(3.5.66)
We choose a particular set of values for the initial data as x(s, 0) = 2s2 ,
y(s, 0) = 2s,
u(s, 0) = 0,
s > 0.
Therefore, the solutions are given by x=
1 2 τ + 2s2 , 2
y = τ + 2s,
u = τ.
(3.5.67)
Eliminating τ and s from (3.5.67) gives the integral surface (u − y)2 + u2 = 2x, or
1 2u = y ± 4x − y 2 2 .
(3.5.68)
The solution surface satisfying the data u = 0 on y 2 = 2x is given by 1 2u = y − 4x − y 2 2 .
(3.5.69)
This represents the solution surface only when y2 < 4x. Thus, the solution does not exist for y 2 > 4x and is not differentiable when y 2 = 4x. We verify that y2 = 4x represents the envelope of the family of characteristics in the (x, t)-plane given by the τ -eliminant of the first two equations in (3.5.67), that is, F (x, y, s) = 2x − (y − 2s)2 − 4s2 = 0.
(3.5.70)
This represents a family of parabolas for different values of the parameter s. Thus, the envelope is obtained by eliminating s from equations ∂F ∂s = 0 and F = 0. This gives y 2 = 4x, which is the envelope of the characteristics for different s, as shown in Figure 3.3.
3.6 Exercises
221
Fig. 3.3 Dotted curve is the envelope of the characteristics.
3.6 Exercises 1. (a) Show that the family of right circular cones whose axis coincides with the z-axis x2 + y 2 = (z − c)2 tan2 α satisfies the first-order, partial differential equation yp − xq = 0. (b) Show that all the surfaces of revolution, z = f (x2 +y 2 ), with the z-axis as the axis of symmetry, where f is an arbitrary function, satisfy the partial differential equation yp − xq = 0. (c) Show that the two-parameter family of curves u − ax − by − ab = 0 satisfies the nonlinear equation xp + yq + pq = u. 2. Find the partial differential equation arising from each of the following surfaces: (a) z = x + y + f (xy), (b) z = f (x − y), (c) z = xy + f (x2 + y 2 ), (d) 2z = (αx + y)2 + β. 3. Find the general solution of each of the following equations: (a) ux = 0, (c) ux + y uy = 0, (e) 2xy ux + (x2 + y 2 )uy = 0, (g) y 2 ux − xy uy = x(u − 2y), (i) y 2 up + u2 xq = −xy 2 ,
(b) aux + b uy = 0, where a, b, are constant, (d) (1 + x2 )ux + uy = 0, (f) (y + u)ux + y uy = x − y, (h) yuy − xux = 1, (j) (y − xu)p + (x + yu)q = x2 + y 2 .
4. Show that the general solution of the equation ux + 2xy2 uy = 0 is given by
222
3 First-Order, Quasi-linear Equations and Method of Characteristics
1 , u = f x2 + y where f is an arbitrary function. Verify by differentiation that u satisfies the original equation. 5. Find the solution of the following Cauchy problems: (a) 3ux + 2uy = 0, with u(x, 0) = sin x, (b) y ux + x uy = 0, with u(0, y) = exp(−y 2 ), (c) x ux + y uy = 2xy, with u = 2 on y = x2 , (d) ux + x uy = 0, with u(0, y) = sin y, (e) y ux + x uy = xy, x ≥ 0, y ≥ 0 with u(0, y) = exp(−y 2 ) for y > 0 and u(x, 0) = exp(−x2 ) for x > 0, (f) ux + x uy = (y − 12 x2 )2 , with u(0, y) = exp(y), (g) x ux + y uy = u + 1, with u(x, y) = x2 on y = x2 , (h) u ux − u uy = u2 + (x + y)2 , with u = 1 on y = 0, (i) x ux + (x + y)uy = u + 1, with u(x, y) = x2 on y = 0. 6. Solve the initial-value problem ut + u ux = 0 with the initial curve
1 2 τ , t = τ, u = τ. 2 7. Find the solution of the Cauchy problem 2 x 2 2xy ux + x + y uy = 0, with u = exp on x + y = 1. x−y x=
8. Solve the following equations: (a) x ux + y uy + z uz = 0, (b) x2 ux + y 2 uy + z(x + y)uz = 0, (c) x(y − z)ux + y(z − x)uy + z(x − y)uz = 0, (d) yz ux − xz uy + xy(x2 + y 2 )uz = 0, (e) x(y 2 − z 2 )ux + y(z 2 − y 2 )uy + z(x2 − y 2 )uz = 0. 9. Solve the equation u x + x uy = y with the Cauchy data (a) u(0, y) = y 2 ,
(b) u(1, y) = 2y.
10. Show that u1 = ex and u2 = e−y are solutions of the nonlinear equation (ux + uy )2 − u2 = 0, but that their sum (ex + e−y ) is not a solution of the equation. 11. Solve the Cauchy problem (y + u)ux + y uy = (x − y),
with u = 1 + x on y = 1.
3.6 Exercises
223
12. Find the integral surfaces of the equation u ux +uy = 1 for each of the following initial data: (a) x(s, 0) = s, y(s, 0) = 2s, u(s, 0) = s, (b) x(s, 0) = s2 , y(s, 0) = 2s, u(s, 0) = s, 2 (c) x(s, 0) = s2 , y(s, 0) = s, u(s, 0) = s. Draw characteristics in each case. 13. Show that the solution of the equation yux − x uy = 0 containing the curve x2 + y 2 = a2 , u = y, does not exist. 14. Solve the following Cauchy problems: (a) x2 ux − y 2 uy = 0, u → ex as y → ∞, (b) y ux + x uy = 0, u = sin x on x2 + y 2 = 1, (c) x ux − y uy = −1 for 0 < x < y, u = 2x on y = 3x, (d) 2x ux + (x + 1)uy = y for x > 0, u = 2y on x = 1, (e) x ux + y uy = x2 + y 2 for x > 0, y > 0, u = x2 on y = 1, (f) y 2 ux + (xy)uy = x, u = x2 when y = 1, (g) x ux + y uy = xy, u = 12 x2 when y = x. 15. Find the solution surface of the equation 2 u − y 2 ux + xy uy + xu = 0, with u = y = x, x > 0. 16. (a) Solve the Cauchy problem ux + uuy = 1,
u(0, y) = ay,
where a is a constant.
(b) Find the solution of the equation in (a) with the data x(s, 0) = 2s, y(s, 0) = s2 , u 0, s2 = s. 17. Solve the following equations: (a) (y + u)ux + (x + u)uy = x + y, (b) x u(u2 + xy)ux − y u(u2 + xy)uy = x4 , (c) (x + y)ux + (x − y)uy = 0, (d) y ux + x uy = xy(x2 − y 2 ), (e) (cy − bz)zx + (az − cx)zy = bx − ay. 18. Solve the equation x zx + y zy = z, and find the curves which satisfy the associated characteristic equations and intersect the helix x + y 2 = a2 , z = b tan−1 ( xy ). 19. Obtain the family of curves which represent the general solution of the partial differential equation (2x − 4y + 3u)ux + (x − 2y − 3u)uy = −3(x − 2y). Determine the particular member of the family which contains the line u = x and y = 0.
224
3 First-Order, Quasi-linear Equations and Method of Characteristics
20. Find the solution of the equation y ux − 2xy uy = 2xu with the condition u(0, y) = y 3 . 21. Obtain the general solution of the equation (x + y + 5z)p + 4zq + (x + y + z) = 0 (p = zx , q = zy ), and find the particular solution which passes through the circle z = 0,
x 2 + y 2 = a2 .
22. Obtain the general solution of the equation 2 z − 2yz − y 2 p + x(y + z)q = x(y − z)
(p = zx , q = zy ).
Find the integral surfaces of this equation passing through (a) the x-axis, (b) the y-axis, and (c) the z-axis. 23. Solve the Cauchy problem (x + y)ux + (x − y)uy = 1,
1 u(1, y) = √ . 2
24. Solve the following Cauchy problems: (a) 3 ux + 2 uy = 0, u(x, 0) = f (x), (b) a ux + b uy = c u, u(x, 0) = f (x), where a, b, c are constants, (c) x ux + y uy = c u, u(x, 0) = f (x), (d) u ux + uy = 1, u(s, 0) = αs, x(s, 0) = s, y(s, 0) = 0. 25. Apply the method of separation of variables u(x, y) = f (x)g(y) to solve the following equations: (a) ux uy = u2 , (b) ux + u = uy , u(x, 0) = 4e−3x , (c) ux + 2uy = 0, u(0, y) = 3e−2y , (d) y 2 u2x + x2 u2y = (xyu)2 , (e) x2 uxy + 9y 2 u = 0, u(x, 0) = exp( x1 ). 26. Use a separable solution u(x, y) = f (x)+g(y) to solve the following equations: (a) u2x + u2y = 1, (b) u2x + u2y = u, 2 2 (c) ux + uy + x = 0, (d) x2 u2x + y 2 u2y = 1, (e) y ux + x uy = 0, u(0, y) = y 2 . 27. Apply v = ln u and then v(x, y) = f (x)+g(y) to solve the following equations: (a) x2 u2x + y 2 u2y = u2 , 2 2 2 2 (b) x2 u √x + y uy = (xyu) . 28. Apply u = v and v(x, y) = f (x) + g(y) to solve the equation x4 u2x + y 2 u2y = 4u.
3.6 Exercises
225
29. Using v = ln u and v = f (x) + g(y), show that the solution of the Cauchy problem 2 u(x, 0) = ex y2 u2x + x2 u2y = (xyu)2 , √ 3 2 2 u(x, y) = exp x + i y . 2
is
30. Consider the eigenvalue problem for the Klein–Gordon equation 1 utt + a2 u, 0 < x < 1, t > 0, c2 u(0, t) = 0 = u(1, t), t > 0, u(x, 0) = 0, and ut (x, 0) = 1, 0 < x < 1. uxx =
(a) Using u(x, t) = X(x)T (t), show that X + λ2 X = 0,
T¨ + a2 + λ2 c2 T = 0,
where −λ2 is a separation constant. (b) Show that the eigenvalues and eigenfunctions are λ = λn = nπ,
Xn (x) = βn sin(nπx),
n = 1, 2, . . . .
(c) Use the Fourier series to obtain the final solution ∞ 2 1 − (−1)n+1 sin(ωn t) sin(nπx), u(x, t) = nπω n n=1 31.
32.
33.
34.
where ωn = a2 + λ2n c2 . Solve the following system of equations with initial data: (a) ut + 3uux = v − x, vt − c vx = 0, u(x, 0) = x, and v(x, 0) = x, (b) ut + 2uux = v − x, vt − c vx = 0, u(x, 0) = x, and v(x, 0) = x. Find the solution of the system of equations with initial data: (a) ut + uux = v e−x , vt − a vx = 0, u(x, 0) = x, and v(x, 0) = ex , (b) ut − 2uux = v − x, vt + c vx = 0, u(x, 0) = x, and v(x, 0) = x. Solve √the following Cauchy problems with initial data: (a) x ux + uuy = −u2 , u(x, 0) = 1, 0 < x < ∞, (b) (x2 u)ux + e−y uy = −u2 , u(x, 0) = 1, 0 < x < ∞. Use ξ(x, t) = x − ct and τ = μt to transform the equation ut + cux = d un ux in the form
∂u ∂u + αun = 0, ∂τ ∂ξ
α = (d/μ).
226
3 First-Order, Quasi-linear Equations and Method of Characteristics
Show that the solution of the equation with u(ξ, 0) = u0 sin kξ is
n u αkun0 τ . u(ξ, τ ) = u0 sin kξ − u0 35. Use the method of characteristics to solve the equations: (a) (x + y)ux + (x − y)uy = 0, (b) ux − (ax + by)uy = 0, (c) x−1 ux + y −1 uy = x2 − y 2 , (d) x sin x ux + y1 uy = u, (e) yux − xuy = 0, u = y when x = a, (f) xux + (x + y)uy = u1 , u = 0, when y = x2 . 36. Solve the first order Fokker–Planck equation with initial condition: ut − xux = u,
u(x, 0) = f (x).
4 First-Order Nonlinear Equations and Their Applications
Physics can’t exist without mathematics which provides it with the only language in which it can speak. Thus, services are continuously exchanged between pure mathematical analysis and physics. It is really remarkable that among works of analysis most useful for physics were those cultivated for their own beauty. In exchange, physics, exposing new problems, is as useful for mathematics as it is a model for an artist. Henri Poincaré Our present analytical methods seem unsuitable for the solution of the important problems arising in connection with nonlinear partial differential equations and, in fact, with virtually all types of nonlinear problems in pure mathematics. The truth of this statement is particularly striking in the field of fluid dynamics. Only the most elementary problems have been solved analytically in this field. . . John Von Neumann
4.1 Introduction First-order, nonlinear, partial differential equations arise in various areas of physical sciences which include geometrical optics, fluid dynamics, and analytical dynamics. An important example of such equations is the Hamilton–Jacobi equation used to describe dynamical systems. Another famous example of the first-order nonlinear equations is the eikonal equation which arises in nonlinear optics and also describes the propagation of wave fronts and discontinuities for acoustic wave equations, Maxwell’s equations, and equations of elastic wave propagation. Evidently, first-order, nonlinear equations play an important role in the development of these diverse areas.
L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_4,
228
4 First-Order Nonlinear Equations and Their Applications
This chapter deals with the theory of the first-order nonlinear equations and their applications. The generalized method of characteristics is developed to solve these nonlinear equations. This is followed by several examples, as well as examples of applications in analytical dynamics, quantum mechanics, and nonlinear optics.
4.2 The Generalized Method of Characteristics The most general, first-order, nonlinear partial differential equation in two independent variables x and y has the form F (x, y, u, p, q) = 0,
(4.2.1)
where u = u(x, y), p = ux , and q = uy . It has been shown in Section 3.3 that the complete solution (integral) of (4.2.1) is a two-parameter family of surfaces of the form f (x, y, u, a, b) = 0,
(4.2.2)
where a and b are parameters. Specifying a space curve, through which the complete integral must pass, generates a solution surface. The equation of the tangent plane at each point (x, y, u) of the solution surface is p(x − ξ) + q(y − η) − (u − ζ) = 0,
(4.2.3)
where (ξ, η, ζ) are running coordinates on the tangent plane. Since the given equation (4.2.1) is a relation between p and q for any solution surface, there is a family of tangent planes corresponding to different values of p and q. We can find an equation that represents the envelope of these planes by considering the intersection of (4.2.3) and the tangent plane at the same point (x, y, u) corresponding to neighboring values p + dp and q + dq of p and q, (p + dp)(x − ξ) + (q + dq)(y − η) − (u − ζ) = 0.
(4.2.4)
Thus, the intersection of (4.2.3) and (4.2.4) leads to the result (x − ξ)dp + (y − η)dq = 0.
(4.2.5)
We can rewrite this equation in terms of quantities specific to the given equation by calculating the differential of (4.2.1) with fixed x, y, and u, that is, Fp dp + Fq dq = 0. Eliminating dp and dq from (4.2.5) and (4.2.6) gives (x − ξ)Fq − (y − η)Fp = 0, or
(4.2.6)
4.2 The Generalized Method of Characteristics
x−ξ y−η = . Fp Fq
229
(4.2.7)
The equation of the surface for which the tangent planes are the envelopes is thus determined by eliminating p and q from (4.2.1), (4.2.3), and (4.2.7). This surface represents a cone for nonlinear equations, called the Monge cone. We next combine (4.2.7) with (4.2.3) to obtain y−η u−ζ x−ξ = = . Fp Fq pFp + qFq
(4.2.8)
Both (x, y, u) and (ξ, η, ζ) lie on the tangent plane, and hence, (4.2.8) represents a line on the tangent plane to the solution surface. Further, x − ξ, y − η, and u − ζ are direction numbers of a line on the tangent plane, and hence, a fixed direction on each tangent plane (for particular values of p and q) is determined by (4.2.8). This direction is determined by the vector (Fp , Fq , pFp + qFq ). As p and q change, these directions, known as the characteristic directions, determine the family of lines (4.2.8) that generate a Monge cone at (x, y, u). To find differential equations for the characteristics as this point moves along a solution surface, we replace x − ξ, y − η, u − ζ by dx, dy, du, respectively, so that (4.2.8) becomes dy du dx = = . Fp Fq pFp + qFq
(4.2.9)
The system (4.2.9) represents equations for the characteristics curves of (4.2.1). As in the case of the quasi-linear equation, equations (4.2.9) depend on the unknown solution u(x, y), but unlike the quasi-linear case, they also depend on p and q. This means that the characteristic curves also depend on the orientation of the tangent planes on the Monge cone at each point. As this geometrical analysis suggests, there is a whole (Monge) cone of characteristics, not just one characteristic, as in the quasilinear case. By equating the ratios in (4.2.9) with dt, we obtain the differential equations for the characteristic curves in the parametric form dx = Fp , dt
dy = Fq , dt
du = pFp + qFq . dt
(4.2.10)
This is a system of three equations for five unknowns x, y, u, p, and q. Thus, for the nonlinear equation (4.2.1), the system (4.2.10) is not a closed set. We need two more equations to close the system, and hence, it is natural to look for equations for dp dq dt and dt . Since p = p[x(t), y(t)] and q = q[x(t), y(t)], we have, by the chain rule and (4.2.10), dx dy dp = px + py = px Fp + py Fq , dt dt dt dx dy dq = qx + qy = qx Fp + qy Fq . dt dt dt Further, u = u(x, y) is a solution of the equation (4.2.1) which gives
(4.2.11) (4.2.12)
230
4 First-Order Nonlinear Equations and Their Applications
dF = Fx + pFu + px Fp + qx Fq = 0, dx dF = Fy + qFu + py Fp + qy Fq = 0. dy
(4.2.13) (4.2.14)
We next use the fact that py = qx , which can be considered an integrability condition for the solution surface, and then utilize (4.2.13), (4.2.14) in (4.2.11), (4.2.12) to obtain dq dp (4.2.15) = −(Fx + pFu ), = −(Fy + qFu ). dt dt Equations (4.2.10) and (4.2.15) form a closed set of five ordinary differential equations for five unknown functions x, y, u, p, and q. They are known as characteristic equations for the nonlinear equation (4.2.1). In principle, these can be solved provided that all five unknown functions are prescribed at t = 0. Usually, as in the case of quasi-linear equations, the Cauchy data are sufficient to specify x, y, and u at t = 0 in terms of some parameter s on an initial curve C. It is also necessary to determine initial values for p and q in terms of s at t = 0. Eliminating dt from (4.2.10) and (4.2.15) gives the Charpit equations dx dy du dp dq = = =− =− . Fp Fq pFp + qFq Fx + pFu Fy + qFu
(4.2.16)
We formulate the Cauchy problem for (4.2.1) to determine the family of curves in the parametric form x = x(t, s),
y = y(t, s),
and u = u(t, s),
(4.2.17)
and the values of p and q as p = p(t, s)
and
q = q(t, s),
(4.2.18)
where t is a running variable along a particular curve and s is a parameter which specifies a member of the family. We seek a solution surface u = u(x, y) with the given initial data at t = 0 as C: x(0, s) = x0 (s),
y(0, s) = y0 (s),
and u(0, s) = u0 (s),
(4.2.19)
and determine the initial conditions for p(t, s) and q(t, s) at t = 0 as p(0, s) = p0 (s)
and
q(0, s) = q0 (s),
(4.2.20)
where p0 (s) and q0 (s) are to be determined. A set of five quantities x0 (s), y0 (s), u0 (s), p0 (s), and q0 (s) cannot all be chosen independently, since we must observe the relation du0 du0 dx0 du0 dy0 = + = p0 (s)x0 (s) + q0 (s)y0 (s), ds dx0 ds dy0 ds
(4.2.21)
4.3 Complete Integrals of Certain Special Nonlinear Equations
231
where p0 (s) and q0 (s) are the initial directions which determine the normal to the solution surface and u0 (s) = u(x0 (s), y0 (s)). The relation (4.2.21) is called the strip condition. The initial curve x0 (s), y0 (s), u0 (s), and the orientations of the tangent planes p0 (s) and q0 (s) are referred to as the initial strip, since the initial conditions can be viewed as the initial curve combined with the initial tangent planes attached. Since p0 (s) and q0 (s) are the initial values of p(t, s) and q(t, s) at t = 0 on the curve C, these values must satisfy the original equation, that is, F (x0 , y0 , u0 , p0 , q0 ) = 0.
(4.2.22)
Evidently, equations (4.2.21) and (4.2.22) determine the initial values p0 (s) and q0 (s). Thus the set of five equations (4.2.10) and (4.2.15) with the initial data (4.2.19), (4.2.20) can be solved to obtain solutions (4.2.17) and (4.2.18). It is necessary to use the first two equations of the system (4.2.17) for finding the parameters t and s in terms of x and y, and the result is substituted into u = u(t, s) to obtain a solution surface u = u(x, y). The sufficient condition for expressing t and s in terms of x and y from the first two equations of the system (4.2.17) is that the Jacobian J is nonzero along the initial strip, that is, ∂(x, y) xt yt Fp Fq = = 0, (4.2.23) = J= xs y s xs y s ∂(t, s) in which the first two equations of the system (4.2.10) are used, where Fp and Fq in (4.2.23) are computed at the initial strip x0 , y0 , u0 , p0 , and q0 . Thus, the above analysis shows that if the Jacobian J = 0 along the initial strip, there exists a unique solution of the equation (4.2.1) passing through the given initial curve (4.2.19). However, in general, the solution is not unique, but it is unique for a fixed determination of the roots p and q of equations (4.2.21) and (4.2.22). However, if J = 0 along the initial strip, then the initial-value problem has a solution if it is a characteristic strip. Indeed, the problem may have infinitely many solutions. The solutions x(t, s), y(t, s), and u(t, s) of the problem for fixed s give a space curve, and p(t, s) and q(t, s) determine a tangent plane with normal vector (p, q, −1) at each point of the space curve. The space curve together with its tangent plane is called the characteristic strip. Similarly, the Charpit equations are known as the characteristic strip equations.
4.3 Complete Integrals of Certain Special Nonlinear Equations Case (i): Nonlinear equations (4.2.1) involving only p and q. Such equations must be of the form F (p, q) = 0. Thus, the Charpit equations (4.2.16) become dx dq dy du dp = = dt. = = = Fp Fq pFp + qFq 0 0
(4.3.1)
232
4 First-Order Nonlinear Equations and Their Applications
The solutions of the last two equations of (4.3.1) are p = const. = a, and q = const. on the characteristic curves. Thus, F (a, q) = 0, which leads to q = f (a). It follows from the first two equations of the system (4.3.1) that du = p dx + q dy, which gives that complete solution by integration as u = ax + f (a)y + b,
(4.3.2)
where b is a constant of integration. Case (ii): Equations not involving independent variables. Such equations are of the form F (u, p, q) = 0.
(4.3.3)
dy du dp dq dx = = = = . Fp Fq pFp + qFq −pFu −qFu
(4.3.4)
Thus, the Charpit equations are
The last equation gives q = ap where a is a constant. Substituting this result into (4.3.3) gives F (u, p, ap) = 0 which can be solved for p = f (u, a), where f is an arbitrary function. Hence, q = af (u, a) and the differential equation for u is du = p dx + q dy = f (u, a) dx + af (u, a) dy. Integrating this equation gives the complete integral u du = x + ay + b, f (u, a)
(4.3.5)
where b is a constant of integration. Case (iii): Separable equations. Such equations are of the form F = G(x, p) − H(y, q) = 0. Thus, the Charpit equations are dx dy du dp dq = = = = = dt. Gp Hq pGp + qHq −Gx Hy It follows from the first and fourth of these equations that Gx dx + Gp dp = 0.
(4.3.6)
4.3 Complete Integrals of Certain Special Nonlinear Equations
233
This gives G(x, p) = const. = a = H(y, q), and hence, p and q can be solved in terms of x and a, and y and a, respectively, to obtain p = g(x, a) and q = h(y, a), so that du = p dx + q dy = g(x, a) dx + h(y, a) dy. This can be integrated to obtain the complete integral u = g(x, a) dx + h(y, a) dy + b,
(4.3.7)
where b is a constant of integration. Case (iv): Clairaut’s equation. A first-order partial differential equation of the form u = px + qy + f (p, q)
(4.3.8)
is called Clairaut’s equation, where f is an arbitrary function of p and q. The associated Charpit equations are dy du dp dx dq = = = = = dt. x + fp y + fq xp + yq + pfp + qfq 0 0 The last two equations imply that p = const. = a and q = const. = b. Substituting these values in (4.3.8) gives the complete integral of (4.3.8) as u = ax + by + f (a, b).
(4.3.9)
This represents the two-parameter family of planes. It can be verified by direct differentiation that (4.3.9) satisfies Clairaut’s equation. Finally, it can also be shown that the characteristic strips are all straight lines. We illustrate the above cases by the following examples. Example 4.3.1. Find the complete integral of the nonlinear equation p2 + qy − u = 0.
(4.3.10)
The associated Charpit equations are dy du dx = Fp = 2p, = Fq = y, = pFp + qFq = 2p2 + qy, dt dt dt dp dq = −(Fx + pFu ) = p, = −(Fy + qFu ) = 0. dt dt The last equation gives q = const. = a. Clearly, the given equation can be solved for p as 1 p = (u − ay) 2 . Also, we have
234
4 First-Order Nonlinear Equations and Their Applications
dx dy du p dx + q dy = = 2 = , 2p y 2p + qy 2p2 + qy and hence,
1
du = p dx + q dy = (u − ay) 2 dx + a dy, or equivalently,
du − a dy √ = dx. u − ay
Integrating this equation gives 1
2(u − ay) 2 = x + b, where b is a constant of integration. Thus, the complete integral is given by u(x, y) =
1 (x + b)2 + ay. 4
(4.3.11)
Example 4.3.2. Solve the Cauchy problem p2 q = 1,
u(x, 0) = x.
(4.3.12)
This equation corresponds to Case (i) with F = p2 q − 1. Hence, dp = 0 and dq = 0 which gives p = const. = a and q = const. = c. Thus, the given equation yields c = a12 . Consequently, the complete integral is u = ax +
1 y + b. a2
(4.3.13)
Using the initial data, we obtain x = ax + b, which gives a = 1 and b = 0. Thus, the solution surface is (4.3.14)
u(x, y) = x + y. Example 4.3.3. Solve the Cauchy problem p2 + q + u = 0,
u(x, 0) = x.
(4.3.15)
The parametric forms of the Charpit equations are dx = 2p, dt
dy = 1, dt
du = 2p2 + q, dt
dp = −p, dt
The parametric forms of the initial data are x(s, 0) = s,
y(s, 0) = 0,
and
u(s, 0) = s.
dq = −q. dt
4.3 Complete Integrals of Certain Special Nonlinear Equations
235
Differentiating u(s, 0) = s with respect to s gives p(s, 0) = 1, and then substituting in the given equation shows that q(s, 0) = −(s + 1). The equations for p and q can be integrated with these initial conditions to obtain p = exp(−t),
q = −(s + 1) exp(−t).
(4.3.16)
Substituting these values of p and q into the first three Charpit equations gives dy = 1, dt
dx = 2 exp(−t), dt
du = 2 exp(−2t) − (s + 1) exp(−t). dt
Integrating these equations with the initial data gives x = (s + 2) − 2 exp(−t),
u = (s + 1) exp(−t) − exp(−2t). (4.3.17)
y = t,
Eliminating t and s gives the solution surface u(x, y) = (x − 1)e−y + e−2y .
(4.3.18)
Example 4.3.4. Find the solution of the initial-value problem p2 x + qy = u,
u(s, 1) = −s.
(4.3.19)
The parametric forms of the Charpit equations are dx = 2px, dt
dy = y, dt
du = 2p2 x + qy, dt
dp = p − p2 , dt
dq = 0. dt
We differentiate u(s, 1) = −s with respect to s to obtain p(s, 0) = −1. Substituting this in the given equation gives q(s, 0) = −2s. Integrating the last equation of the Charpit system gives q(s, t) = −2s. The equation for p can be integrated to obtain Ap(1 − p)−1 = exp(t), where A is a constant of integration. Using the initial data p(s, 0) = −1, we find that A = −2, and hence, et p(s, t) = t . (4.3.20) (e − 2) Substituting this solution for p in the first equation of the Charpit set and integrating with the condition x(s, 0) = s, we obtain 2 x(s, t) = s et − 2 .
(4.3.21)
The equation for y can be solved with y(s, 0) = 1 and the solution is y(s, 0) = exp(t), so that the equations of characteristics are x = s(y − 2)2 . They will intersect at (0, 2). Finally, the equation for u can be expressed in terms of t and s as du = 2s e2t − et . dt
236
4 First-Order Nonlinear Equations and Their Applications
Integrating this equation with u(s, 0) = −s gives the solution u(s, t) = s exp(t) exp(t) − 2 .
(4.3.22)
In terms of x and y, the solution takes the form u(x, y) =
xy . y−2
(4.3.23)
This shows that the solution u(x, y) is singular at y = 2. Example 4.3.5. Solve the initial-value problem p2 − 3q2 − u = 0 with u(x, 0) = x2 .
(4.3.24)
The characteristic strip equations are dx dy du = Fp = 2p, = Fq = −6q, = 2 p2 − 3q 2 = 2u, dt dt dt dp dq = −(Fx + pFu ) = p, = −(Fy + qFu ) = q. dt dt
(4.3.25) (4.3.26)
The initial data in the parametric form are given by u(0, s) = u0 (s) = s2 , (4.3.27) and p0 (s) and q0 (s) are the solutions of equations (4.2.21) and (4.2.22) which become (4.3.28) p(0, s) = p0 (s) = 2s, and p20 − 3q02 = s2 , x(0, s) = x0 (s) = s,
y(0, s) = y0 (s) = 0,
and hence, q0 (s) = ∓s.
(4.3.29)
Next we use p0 (s) = 2s and q0 (s) = −s to solve (4.3.26). It turns out that p(t, s) = 2set
and q(t, s) = −set .
(4.3.30)
Substituting these results in (4.3.25) and solving the resulting equations with the initial data (4.3.27) gives x(t, s) = 4s et − 1 + s, y(t, s) = 6s et − 1 , and u(t, s) = s2 e2t . (4.3.31) We use the first two results in (4.3.31) to find s and t in terms of x and y as s=
1 (3x − 2y) 3
and
exp(t) =
3 (2x − y) . 2 (3x − 2y)
(4.3.32)
Substituting these results into the third equation of the system (4.3.31) gives the solution surface
4.3 Complete Integrals of Certain Special Nonlinear Equations
2 y u(x, y) = x − . 2
237
(4.3.33)
If we use the initial data p0 (s) = 2s and q0 (s) = s and solve the problem in a similar way, we obtain a different solution surface 2 y . u(x, y) = x + 2
(4.3.34)
Example 4.3.6. Solve the Clairaut equation 1 z = px + qy + f (p, q) = px + qy + 1 + p2 + q 2 2 ,
(4.3.35)
with the initial data x(0, s) = x0 (s) = a cos s,
y(0, s) = y0 (s) = a sin s,
z(0, s) = z0 (s) = 0.
(4.3.36)
The characteristic strip equations are p dx = x+ , dt f dp = 0, and dt
dy q =y+ , dt f dq = 0. dt
dz 1 =z− , dt f
(4.3.37) (4.3.38)
We have to find the initial data p(0, s) = p0 (s) and q(0, s) = q0 (s) which must satisfy the original equation and the strip condition, that is, 1 (4.3.39) ap0 (s) cos s + aq0 (s) sin s + 1 + p20 + q02 2 = 0, −ap0 (s) sin s + aq0 (s) cos s = 0 (a = 0). (4.3.40) Clearly, equation (4.3.40) is satisfied if p0 (s) = a0 (s) cos s
and
q0 (s) = a0 (s) sin s.
(4.3.41)
Putting these results into (4.3.39) gives (a2 − 1)a20 (s) = 1 which leads to − 1 a0 (s) = ± a2 − 1 2 = const. (a > 1). (4.3.42) We take the negative sign for a0 (s) to solve equations (4.3.37) and (4.3.38). In fact, the latter equation gives the solutions p(t, s) = const. = −(a2 − 1)− 2 cos s, 1
q(t, s) = const. = −(a2 − 1)− 2 sin s. 1
Then,
− 1 1 f = 1 + p2 + q 2 2 = a a2 − 1 2 ,
which is used to rewrite the first two equations in (4.3.37) as
(4.3.43ab)
(4.3.44)
238
4 First-Order Nonlinear Equations and Their Applications
dx p cos s =x+ =x− , dt f a
dy sin s =y− . dt a
(4.3.45)
The solution of the first equation in (4.3.45) with x(0, s) = a cos s is given by 2
a −1 1 1 t 1 e + cos s = et + 2 cos s. x(t, s) = a− a a a a −1 (4.3.46) Similarly,
2 1 t 1 1 a −1 t y(t, s) = a− e + sin s = e + 2 sin s, (4.3.47) a a a a −1 1 (a2 − 1) 2 t (4.3.48) z(t, s) = − e −1 . a Or equivalently, √ a 1 a2 − 1 t z−√ e + 2 . =− a a −1 a2 − 1 Thus, the equation of the characteristic strip is
2 a x2 + y 2 = a2 − 1 z − √ . a2 − 1
(4.3.49)
(4.3.50)
This represents a cone. In general, any solution of the Clairaut equation represents a developable surface, that is, a ruled surface which can be deformed into a plane without stretching or tearing.
4.4 The Hamilton–Jacobi Equation and Its Applications In analytical dynamics, the Hamilton principle function S(qi , t) characterizes the dynamical system and satisfies the celebrated Hamilton–Jacobi equation ∂S ∂S = 0, i = 1, 2, . . . , n, (4.4.1) + H qi , t, ∂t ∂qi where qi are the generalized coordinates, t is the time variable, and H is the Hamiltonian of a dynamical system. This equation is a first-order, nonlinear, partial differential equation with n + 1 independent variables (qi , t), and it plays a fundamental role in the development of analytical dynamics. It follows from (4.4.1) that S is equal to the time integral of the Lagrangian L. We have dS ∂S ∂S = −H + q˙i pi = L, = + q˙i dt ∂t ∂qi where pi =
∂S ∂qi .
(4.4.2)
4.4 The Hamilton–Jacobi Equation and Its Applications
Integrating (4.4.2) from t1 and t2 gives t2 L dt. S=
239
(4.4.3)
t1
This shows that S is a functional which satisfies the Hamilton principle. Hamilton realized that S is a solution of the Hamilton–Jacobi equation. ∂S ∂u = ∂q = pi , and p = St , Using the conventional notation S(qi , t) = u(qi , t), ∂q i i we rewrite equation (4.4.1) in the form F (qi , t, pi , p) = p + H(qi , t, pi ) = 0,
(4.4.4)
where the dependent variable u(qi , t) does not occur in the equation. In terms of the parameter τ , the Charpit equations associated with (4.4.4) are given by dqi ∂H = Fpi = , dτ ∂pi dt = Fp = 1, dτ n n ∂H du = pi Fpi + pFp = pi + p, dτ ∂pi i=1 i=1 ∂H dpi = −(Fqi + pi Fu ) = − , dτ ∂qi ∂H dp = −(Ft + pFu ) = − . dτ ∂t
(4.4.5) (4.4.6) (4.4.7) (4.4.8) (4.4.9)
The second equation (4.4.6) with the given initial condition t(τ ) = 0 at τ = 0 gives t = τ . Thus, the independent time variable t can be used as the parameter of the characteristics. Thus, the above system of equations reduces to the form dqi ∂H , = dt ∂pi dpi ∂H =− , dt ∂qi n du ∂H = pi + p, dt ∂pi i=1 ∂H dp =− . dt ∂t
(4.4.10) (4.4.11) (4.4.12) (4.4.13)
The first two equations (4.4.10), (4.4.11) constitute a set of 2n coupled, first-order, ordinary differential equations, precisely the Hamilton canonical equations of motion, which reflect symmetry except for a negative sign. The solutions of these equations represent the characteristics of the Hamilton–Jacobi equation. They also represent the generalized coordinates and generalized momenta of a dynamical system
240
4 First-Order Nonlinear Equations and Their Applications
whose Hamiltonian is H(qi , pi , t). The last equation (4.4.13) gives p = −H which is used to rewrite (4.4.12) in the form du ∂H = pi − H. dt ∂pi
(4.4.14)
In principle, the Hamilton system of 2n equations can be solved for qi (t) and ∂H ) and H in (4.4.12), the right-hand side pi (t). If we substitute these solutions in ( ∂p i of (4.4.12) gives a known function of t, and then u(qi , t) can be found by integration. A similar argument can be used to find p by integrating (4.4.13). Thus, the upshot of this analysis is that the characteristics of the Hamilton–Jacobi equation are the solutions of the Hamilton equations. By using Hamilton’s equations (4.4.10), (4.4.11), the equations of motion for any canonical function F (qi , pi , t) can be expressed in the form n ∂F dF ∂F ∂F = q˙i + p˙i + dt ∂q ∂p ∂t i i i=1 n ∂F ∂F ∂H ∂F ∂H + = − ∂qi ∂pi ∂pi ∂qi ∂t i=1 = {F, H} +
∂F , ∂t
(4.4.15)
where {F, H} is called the Poisson bracket of two functions F and H. If the canonical function F does not explicitly depend on time t, then Ft = 0 and hence (4.4.15) becomes dF = {F, H}. (4.4.16) dt In addition, if {F, H} = 0, then F is a constant of motion. In fact, equation (4.4.16) really includes the Hamilton equations, which can easily be verified by setting F = qi and F = pi . Further, if F = H, (4.4.16) implies that H is constant. More generally, the Poisson bracket of any two functions F and G is defined by ∂F ∂G ∂F ∂G . (4.4.17) {F, G} = − ∂qi ∂pi ∂pi ∂qi i Obviously, the Poisson bracket is antisymmetric. It also readily follows from the definition of the Poisson bracket that {qi , pj } = δij , {qi , qj } = 0 = {pi , pj },
(4.4.18) (4.4.19ab)
where δij is the Kronecker delta notation. Results (4.4.18), (4.4.19ab) represent the fundamental Poisson brackets for the canonically conjugate variables qi and pi . It is important to point out that the sum in a Poisson bracket is taken over all the independent variables. Partial derivatives of only two dependent variables occur
4.4 The Hamilton–Jacobi Equation and Its Applications
241
in a given Poisson bracket, and these two variables are used in the notation for that bracket. To treat time t similarly to the other independent variables, we replace t by qn+1 and write (4.4.4) in the form G(qi , qn+1 , pi , pn+1 ) = pn+1 + H(qi , t, pi ).
(4.4.20)
This allows us to write the Hamilton equations in the form ∂G dqi = , dt ∂pi
∂G dpi =− , dt ∂qi
(4.4.21ab)
where G = G(qi , pi ) and i = 1, 2, . . . , n + 1. Evidently, if G is given, equations (4.4.21ab) represent a system of (2n + 2) first-order, ordinary differential equations for the unknown functions qi and pi . The main problem is to find a suitable transformation under which the symmetric form of the Hamilton equations is preserved. Such a transformation is called a canonical transformation, defined as follows. A transformation of the form qi = qi (˜ qi , p˜i ),
pi = pi (˜ qi , p˜i ),
(4.4.22ab)
˜ qi , p˜i ) such where i = 1, 2, . . . , n+1, is called canonical if there exists a function G(˜ that equations (4.4.21ab) transform into ˜ ∂G d˜ qi = , dt ∂ p˜i
˜ ∂G d˜ pi =− . dt ∂ q˜i
(4.4.23ab)
There are other mathematical expressions invariant under canonical transformations of the form (4.4.22ab). One such set is the integral invariants of Poincaré. A theorem of Poincaré states that the integral J1 = dqi dpi (4.4.24) S
i
is invariant under the canonical transformation, where S indicates that the integral is to be evaluated over any arbitrary two-dimensional surface in the phase space formed by coordinates q1 , q2 , . . . , qn , and p1 , p2 , . . . , pn . It is noted that the position of a point on any two-dimensional surface is completely specified by not more than two parameters. We assume that u and v are two such parameters appropriate to the surface S, so that qi = qi (u, v) and pi = pi (u, v) on this surface. It is well known that the elementary area dqi dpi transforms to the element of area du dv according to the relation ∂(qi , pi ) du dv, (4.4.25) dqi dpi = ∂(u, v) where the Jacobian determinant is given by ∂q ∂(qi , pi ) ∂ui = ∂qi ∂(u, v) ∂v
∂pi ∂u ∂pi ∂v
= 0.
(4.4.26)
242
4 First-Order Nonlinear Equations and Their Applications
The integral J1 has the same value for all canonical coordinates, that is, dqi dpi = d˜ qk d˜ pk , S
S
i
(4.4.27)
k
which can also be expressed as ∂(qi , pi ) ∂(˜ qk , p˜k ) du dv = du dv. ∂(u, v) ∂(u, v) S i S
(4.4.28)
k
Since the region of integration is arbitrary, the two integrals in (4.4.28) are equal, provided that the two integrands are equal, that is, ∂(qi , pi ) ∂(u, v)
i
=
∂(˜ qk , p˜k ) k
∂(u, v)
(4.4.29)
.
This means that the sum of the Jacobian determinants is invariant. More explicitly, (4.4.29) can be written as ∂qi ∂pi ∂ q˜k ∂ p˜k ∂pi ∂qi ∂ p˜k ∂ q˜k = . (4.4.30) · − · · − · ∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v i k
Each side of this equation is in the form of what is called the Lagrange bracket of two independent variables u and v defined by ∂qi ∂pi ∂pi ∂qi (u, v) = . (4.4.31) · − · ∂u ∂v ∂u ∂v i More generally, if Fk and Gk , k = 1, 2, . . . , n are a set of functions of a number of independent variables, and if u and v are any two of these variables, then the Lagrange bracket is defined by ∂Fk ∂Gk ∂Fk ∂Gk (u, v) = . (4.4.32) · − · ∂u ∂v ∂u ∂v k
Obviously, the Lagrange bracket is antisymmetric. We now consider the Lagrange bracket of the generalized coordinates ∂qk ∂pk ∂qk ∂pk = 0. (qi , qj ) = · − · ∂qi ∂qj ∂qj ∂qi
(4.4.33)
k
Since the q’s and p’s are independent coordinates, Similarly, we can show that (pi , pj ) = 0.
∂pk ∂qi
= 0 and
∂pk ∂qj
= 0.
(4.4.34)
4.4 The Hamilton–Jacobi Equation and Its Applications
Finally, it can be shown that ∂qk ∂pk ∂qk ∂pk = δij . · − · (qi , pj ) = ∂qi ∂pj ∂pj ∂qi
243
(4.4.35)
k
The second term in the above sum vanishes because the q’s and p’s are independent, but the first term is not zero because ∂qk = δki , ∂qi and hence, (qi , pj ) =
and
∂pk = δkj , ∂pj
(4.4.36ab)
δjk δki = δij .
k
Thus, (4.4.33)–(4.4.35) represent the Lagrange brackets of canonical variables and are often referred to as the fundamental Lagrange brackets. In contrast to Poisson brackets, the sum in a Lagrange bracket is taken over all the dependent variables. Partial derivatives with respect to two of the independent variables occur in a given Lagrange bracket, and these two independent variables are utilized in the notation for the bracket. The following observations are in order. First, the condition that the transformation be canonical can be expressed either in terms of Poisson brackets or in terms of Lagrange brackets. Second, there exists a mathematical relation between the Poisson and Lagrange brackets which can be found in Goldstein (1965). Third, both Poisson and Lagrange brackets are found useful in the transition from classical mechanics to quantum mechanics. Fourth, for more information on Hamilton–Jacobi equations, see Qiao (2001, 2002). Example 4.4.1 (The Hamilton–Jacobi Equation for a Single Particle). We consider a conservative dynamical system, and write S = u − Et where u is independent of time t and E is an arbitrary constant. We see that S is a solution of the Hamilton– Jacobi equation (4.4.1) if u satisfies the time-independent Hamilton–Jacobi equation in the form ∂u = E, (4.4.37) H qi , ∂qi where u is called Hamilton’s characteristic function. We now derive the Hamilton–Jacobi equation for a single particle of mass m moving under the influence of a conservative force field. The Hamiltonian of this system is given by H =T +V =
1 2 p + p22 + p23 + V (x, y, z), 2m 1
(4.4.38)
where V (x, y, z) is the potential energy of the particle and pi =
∂S ∂u = , ∂qi ∂qi
i = 1, 2, 3.
(4.4.39)
244
4 First-Order Nonlinear Equations and Their Applications
We may use Cartesian coordinates as the generalized coordinates, that is, q1 = x, q2 = y, and q3 = z, to reduce the Hamilton–Jacobi equation (4.4.38) in the form 1 2m
∂u ∂x
2
+
∂u ∂y
2
+
∂u ∂z
2 + V (x, y, z) = E,
(4.4.40)
or equivalently, u2x + u2y + u2z = f (x, y, z),
(4.4.41)
where f (x, y, z) = 2m(E − V ). This is a first-order nonlinear equation. In the next section, we discuss the solution of this equation in the context of nonlinear geometrical optics. Example 4.4.2 (Simple Harmonic Oscillator). The Hamilton–Jacobi equation can readily be applied to solve the problem of a simple harmonic oscillator. In terms of generalized coordinates (q, p), the Hamiltonian of this conservative problem is given by k p2 + q2, H =T +V =E = (4.4.42) 2m 2 where m is the mass and k is a positive constant. , the Hamilton–Jacobi equation (4.4.1) becomes Setting p = ∂S ∂q 2 ∂S 1 ∂S k + + q 2 = 0. ∂t 2m ∂q 2
(4.4.43)
For this conservative system, the time-independent part of S can be separated by introducing a function u so that S(q, t) = u(q) − Et. Consequently, (4.4.43) reduces to the form 2 k 1 ∂u + q 2 = E. (4.4.44) 2m ∂q 2 Integrating this equation gives √ 2E u = mk − q 2 dq = S + Et. k
(4.4.45)
Since S is independent of E, we differentiate (4.4.45) with respect to E to obtain ∂u m m k dq −1 =− q + τ = t, (4.4.46) = cos ∂E k k 2m ( 2E ) − q 2 k
where τ is a constant of integration. Thus, solving (4.4.46) for q yields 2E k . ω2 = cos ω(t − τ ) q(t) = k m
(4.4.47)
4.4 The Hamilton–Jacobi Equation and Its Applications
The momentum conjugate to q is given by √ p = mq˙ = − 2mE sin ω(t − τ ) .
245
(4.4.48)
A simple calculation of L = T − V can be used to verify that S is equal to the time integral of L. Example 4.4.3 (Jacobi’s Principle of Least Action). For a conservative dynamical system (H = E), the principle of least action is given by t2 m 2T dt T = s˙ 2 0 = δA = δ 2 t s2 s2 √ 1 2mT ds = δ 2m(E − V ) ds. (4.4.49) =δ s1
s1
This is known as the Jacobi principle of least action, and it is concerned with the path of the system in the configuration space, rather than the motion in time. Indeed, the time does not appear in the integrand of the integral in (4.4.49) because V = V (q) does not depend on time. This principle in the same form as (4.4.49) is also valid for a conservative system of particles. Example 4.4.4 (Wave Propagation in Continuous Media). The Hamilton–Jacobi theory can also be applied to problems of wave propagation in continuous media. In a conservative dynamical system, the Hamiltonian is a constant of motion and is identified with the total energy. For an n-dimensional configuration space where q = (q1 , q2 , . . . , qn ), the Hamilton function S can be expressed in terms of the characteristic function u as S(q, t) = u(q) − Et, (4.4.50) where u is independent of time t, and the surface of constant u is fixed in the configuration space. However, the surface of constant S moves with time according to (4.4.50). We suppose that at some time t the surface of constant S corresponds to the surface of constant u so that at time t + dt that surface coincides with the surface for which u = S + Edt. Obviously, during the small time interval dt, the surface of constant S moves to a new surface u + du. Thus, the family of surfaces, S = const., can be interpreted as the family of wave fronts propagating in the space. The outward normal vector at each point on the surface S = const. represents the direction of the phase (wave) velocity cp = ds dt , where ds is the distance normal to the surface, and S = a constant that moves from u to a new position U +dU in time dt. The definition of du and (4.4.50) reveals that du = ∇u · c = E dt
(4.4.51)
ds E = . dt |∇u|
(4.4.52)
so that the phase velocity is cp =
246
4 First-Order Nonlinear Equations and Their Applications
For a single particle of mass m in the configuration space q = (x, y, z), the Hamiltonian is H = T + V = E so that the Hamilton–Jacobi equation for S is E−V =
2 1 1 ∂u = |∇u|2 , 2m ∂qi 2m
so that |∇u|2 = 2m(E − V ).
(4.4.53)
This result can be used to rewrite the phase speed (4.4.52) in the form c=
E E E = = . |∇u| p 2m(E − V )
In an n-dimensional configuration space, pi =
∂u ∂qi
p = ∇u.
(4.4.54)
which is, in vector notation, (4.4.55)
Evidently, ∇u determines the normal to the surface of the constant S or u, and hence, represents the direction of wave propagation. It also follows from (4.4.54) that pc = E = const., which asserts that when the surfaces move slower, particles move faster, and vice versa.
4.5 Applications to Nonlinear Optics Light waves are electromagnetic waves which are transverse waves that can be described by both scalar and vector potentials. For simplicity, we consider the scalar wave equation for the electromagnetic potential φ ∇2 φ −
n2 φtt = 0, c20
(4.5.1)
where c0 is the speed of light in a vacuum, n = cc0 is called the refractive index of the medium, and c is the speed of light in an optical medium. In general, n depends on the optical density of the medium and is a function of space variables. For constant refractive index n, equation (4.5.1) admits a plane wave solution (4.5.2) φ = A0 exp i(k · x − ωt) , where A0 is a complex wave amplitude, provided the following dispersion relation holds: 2 n2 ω 2 k + l2 + m2 = 2 , k = (k, l, m). (4.5.3) c0 In the case of one-dimensional wave propagation problems, the wave number k = is given by
2π λ
4.5 Applications to Nonlinear Optics
k=
nω ω = . c0 c
247
(4.5.4)
In general, n depends on x, and the wave form is likely to be changed so that the exact plane wave solutions may not be possible. We adopt a geometrical optics approximation where n varies slowly with distance which is assumed to be the order of the wavelength λ. In geometrical optics, λ is small (or k is large) where (4.5.4) holds. Thus, the geometrical optics approximation deals with a small wavelength or high frequency of waves. It is then natural to seek a nearly plane wave solution in the form (4.5.5) φ = exp A(x) + ik0 ψ(x) − c0 t , where the wave amplitude A and the phase ψ of the wave are slowly varying functions of x, and k0 is the wavenumber in the vacuum (n = 1) with k0 = nω c0 . Application of the gradient operator to (4.5.5) gives ∇φ = φ∇(A + ik0 ψ). 2 ∇2 φ = φ ∇2 (A + ik0 ψ) + ∇(A + ik0 ψ) = φ ∇2 A + ik0 ∇2 ψ + (∇A)2 − k02 (∇ψ)2 + 2ik0 ∇A · ∇ψ .
(4.5.6) (4.5.7)
Consequently, the wave equation (4.5.1) becomes ∇2 A + (∇A)2 + k02 n2 − (∇ψ)2 φ + ik0 2∇A · ∇ψ + ∇2 ψ φ = 0. (4.5.8) Both A and ψ are real, and hence, the real and imaginary parts of (4.5.8) must vanish so that 2∇A · ∇ψ + ∇2 ψ = 0, ∇2 A + (∇A)2 + k02 n2 − (∇ψ)2 = 0.
(4.5.9) (4.5.10)
The geometrical optics approximation has no effect on equation (4.5.9), but equation (4.5.10) is modified in the limit as k0 → ∞ so that it becomes (∇ψ)2 = n2 .
(4.5.11)
This is known as the eikonal equation in optics, and the phase function ψ is called the eikonal (eikon is a Greek word for image or figure). This is a first-order nonlinear partial differential equation which plays a very important role in optics and in wave propagation in continuous media. Physically, equation (4.5.11) determines a surface of constant phase as wave fronts. There is a remarkable correspondence between the eikonal equation (4.5.11) in optics and the Hamilton–Jacobi equation (4.4.40) in classical mechanics, where n in the former equation plays a role similar to that of 2m(E − V ) in the latter. Thus, there is a striking similarity between the eikonal ψ and the characteristic function u in classical mechanics. It is also evident from the Hamilton–Jacobi theory that classical mechanics may be regarded as the small wavelength (or high frequency) limit of the geometrical optics. This relationship
248
4 First-Order Nonlinear Equations and Their Applications
between classical mechanics and geometrical optics can be reconfirmed from the fact that Fermat’s principle of least time follows from Jacobi’s principle of least action as seen below: s2 s2 s2 s2 2m(E − V ) ds = δ n ds = δ (nc) dt = δ c0 dt. 0=δ s1
s1
s1
s1
(4.5.12) Example 4.5.1 (Eikonal Equation in Nonlinear Geometrical Optics). In optics, light waves propagate in a medium along rays with a given speed c = c(x, y, z) depending on space variables. The surfaces of constant optical phase, x0 (τ ) = τ , y0 (τ ) = τ , and u(x, y, z) = const., are called the wave fronts, which are orthogonal to the light rays. The function u(x, y, z) satisfies the classical, nonlinear, first-order equation u2x + u2y + u2z = n2 (x, y, z),
(4.5.13)
where n = ( cc0 ) is called the index of refraction. In two space dimensions, the eikonal equation (4.5.13) takes the form u2x + u2y = n2 (x, y).
(4.5.14)
This describes the propagation of cylindrical waves in optics, acoustics, elasticity, and electromagnetic theory. The classical and simplest problem of wave propagation deals with shallow water waves where wave crests or troughs, or any curve of constant phase, are the level curves expressed as u(x, y) − at = const.,
(4.5.15)
with t as the time variable. As time progresses, waves propagate in the (x, y)-plane. When n is a constant, we can apply Case (i) in Section 4.3 to find the general solution of (4.5.14). It follows from (4.3.2) that the complete solution of (4.5.14) can be written as u(x, y) = anx + n 1 − a2 y + b, (4.5.16) where a and n are constants. Or equivalently, u(x, y) = nx cos θ + ny sin θ + b,
(4.5.17)
where a = cos θ. With n(x, y) = const. = n, we solve (4.5.14) as the initial-value problem for x(s, τ ), y(s, τ ), u(s, τ ), p(s, τ ), and q(s, τ ) with the Cauchy data x(0, τ ) = x0 (τ ), p(0, τ ) = p0 (τ ), where s and τ are parameters.
y(0, τ ) = y0 (τ ), and
q(0, τ ) = q0 (τ ),
u(0, τ ) = u0 (τ ),
(4.5.18) (4.5.19)
4.5 Applications to Nonlinear Optics
249
According to the general theory of first-order equations, the Charpit equations associated with (4.5.14) are dx = 2p, ds dp = 0, ds
dy = 2q, ds
du = 2 p2 + q 2 , ds
dq = 0. ds
(4.5.20)
Evidently, the last two equations in (4.5.20) give p(s, τ ) and q(s, τ ) which are constant on the characteristics, that is, p(s, τ ) = const. = p0 (t)
and
q(s, τ ) = const. = q0 (t).
(4.5.21)
Since p and q are constant for fixed τ , the first two equations in (4.5.20) imply that characteristics (light rays) are straight lines which can be obtained by direct integration in the form x(s, τ ) = 2p0 s + x0 (τ ),
y(s, τ ) = 2q0 s + y0 (τ ).
By integration, the third equation in (4.5.20) gives u(s, τ ) = 2 p20 + q02 s + u0 (τ ) = 2n2 s + u0 (τ ).
(4.5.22)
(4.5.23)
Along the characteristics (or light rays), equation (4.5.23) has the same form as (4.5.15) with s playing the role of a scaled-time variable and where u0 (τ ) is independent of time like variable s. Eliminating p0 , q0 , and s from (4.5.22) and (4.5.23) gives the solution (u − u0 )2 = n2 (x − x0 )2 + (y − y0 )2 . (4.5.24) This represents a cone with its vertex at (x0 , y0 , u0 ). Or equivalently, taking the positive square root, we have 1 u(x, y) = u0 + n (x − x0 )2 + (y − y0 )2 2 .
(4.5.25)
This is a three-parameter family of characteristic strips of (4.5.14), and hence, it represents the solution of the eikonal equation (4.5.14). However, the solution (4.5.25) satisfies (4.5.14) everywhere except at the initial point (x0 , y0 ), where ux and uy are singular. Thus, (4.5.25) is called the singular solution which determines a cone with its vertex at (x0 , y0 , u0 ), identical with the Monge cone through that point. Physically, the level curves u(x, y) = const. characterize the wave fronts, where ∇u = (ux , uy ) = (p, q) represents the normal vector to the wave fronts. It follows from (4.5.20) that 2(p, q) represents the tangent vector to the characteristics or light rays. Thus, the light rays are normal to the wave fronts. This is true even for a variable index of refraction.
250
4 First-Order Nonlinear Equations and Their Applications
Now, we discuss several special solutions of eikonal equation (4.5.14) with constant n and the following three sets of initial data: (i) x0 (τ ) = 0,
and
y0 (τ ) = 0,
u0 (τ ) = 0.
In this case, solutions (4.5.22) and (4.5.23) reduce to the form x = 2p0 s,
y = 2q0 s,
and u = 2n2 s.
Eliminating s from the first two results gives x2 + y 2 = 4 p20 + q02 s2 = 4n2 s2 .
(4.5.26)
(4.5.27)
Therefore, the solution is 1 u(x, y) = n x2 + y 2 2 .
(4.5.28)
The level curves representing the wave fronts are u(x, y) = const., which represents a family of concentric circles. The strip condition (4.2.21) is automatically satisfied, the given equation leads to p20 + q02 = n2 , and hence, p0 and q0 are constant. Thus, the characteristics representing the rays are given by a family of straight lines q0 x − p0 y = 0 which passes through the origin. This example also confirms the fact that rays are normal to the wave fronts. (ii) x0 (τ ) = τ,
y0 (τ ) = τ,
u0 (τ ) = aτ.
It follows from equation (4.5.14) that p0 (τ ) = n cos θ
and
q0 (τ ) = n sin θ,
for some real values of θ. In order to satisfy the strip condition (4.2.21) in which s is replaced by τ , the result n(cos θ + sin θ) = a (4.5.29) must hold for some values of θ. Consequently, solutions (4.5.22) and (4.5.23) become x(s, τ ) = 2ns cos θ + τ, y(s, τ ) = 2ns sin θ + τ, 2 u(s, τ ) = 2n s + nτ (cos θ + sin θ).
(4.5.30) (4.5.31)
Eliminating s and τ from (4.5.30) gives s=
(x − y) 2n(cos θ − sin θ)
and τ =
(y cos θ − x sin θ) . (cos θ − sin θ)
(4.5.32)
Substituting the values for s and τ in (4.5.31) leads to the solution u(x, y) = n(x cos θ + y sin θ).
(4.5.33)
4.5 Applications to Nonlinear Optics
251
In particular, when a = n, then cos θ + sin θ = 1, which implies that either θ = 0 or θ = π2 . Therefore, solution (4.5.33) leads to two continuously differentiable solutions given by or ny. cos τ (iii) x0 (τ ) = , 1 − cos τ
u(x, y) = nx
y0 (τ ) =
sin τ , 1 − cos τ
(4.5.34) n cos τ and u0 (τ ) = . 1 − cos τ (4.5.35)
The first two initial conditions in (4.5.35) determine the initial curve y2 = 2x+1, which is a parabola with its vertex at (− 12 , 0) and its focus at (0, 0). To determine the initial data p0 (τ )and q0 (τ ), we need to satisfy the given equation (4.5.14) and the strip condition as p20 (τ ) + q02 (τ ) = n2 , p0 (τ )x0 (τ ) + q0 (τ )y0 (τ ) = u0 (τ ).
(4.5.36) (4.5.37)
These equations give two sets of solutions: (a) p0 = n, q0 = 0;
(b) p0 = n cos τ, q0 = n sin τ.
For case (a), the solution can be obtained from (4.5.22) and (4.5.23) as x(s, τ ) = 2ns + x0 (τ ),
y(s, τ ) = y0 (τ ),
u(s, τ ) = 2n2 s + u0 (τ ). (4.5.38)
From the first and the third results with the data (4.5.35), we obtain the solution (4.5.39)
u(x, y) = nx.
The level curves u = const. give x = const., which is a family of straight lines parallel to the y-axis as shown in Figure 4.1. For case (b), the solutions are found from (4.5.22) and (4.5.23) as x = 2ns cos τ + x0 (τ ),
y = 2ns sin τ + y0 (τ ),
u(s, τ ) = 2n2 s + u0 (τ ).
After some algebraic simplification, we obtain the solution 1 u(x, y) = n x2 + y 2 2 − 1 .
(4.5.40)
Thus, the wave fronts represented by the level curves u(x, y) = const. are x 2 + y 2 = a2 . They represent concentric circles as shown in Figure 4.1.
(4.5.41)
252
4 First-Order Nonlinear Equations and Their Applications
Fig. 4.1 The level curves (wave fronts) for solutions (4.5.39) and (4.5.41).
Example 4.5.2 (Eikonal Equation with a Point Source). We consider the initial-value problem for the eikonal equation with a point source which moves along a straight line in the (x, t)-plane with constant speed v. Without loss of generality, we assume that motion is set up along the x-axis and passes through the origin when the eikonal u = 0. We solve the equation 1 (4.5.42) p2 + q 2 = 2 , c with the Cauchy data x(0, τ ) = x0 (τ ) = τ,
y(0, τ ) = y0 (τ ) = 0,
τ , v (4.5.43)
u(0, τ ) = u0 (τ ) =
where c is a constant. The conditions (4.5.36), (4.5.37) give 1 c2 1 p20 (τ ) = , v
p20 (τ ) + q02 (τ ) =
and hence, q02 (τ ) = ±
and
1 1 2 M − 1 2, v
(4.5.44a) (4.5.44b)
(4.5.45)
where M = vc is the Mach number. The condition (4.5.44a) is satisfied if we set p0 (τ ) = cosc θ and q0 (τ ) = sinc θ , 1 where the new parameter θ is defined by θ = cos−1 ( M ) which has real solutions for θ in (0, 2π) provided the Mach number M > 1. Thus, the solutions of the present problem can be obtained from (4.5.22) and (4.5.23) as 2s + τ, v 2s τ u(x, τ ) = 2 + , c v x(s, τ ) =
1 2s 2s 2 sin θ = M −1 2, c cM 1 1 2 p(s, τ ) = M c, q(s, τ ) = M − 1 2. cM
y(s, τ ) =
(4.5.46) (4.5.47)
4.5 Applications to Nonlinear Optics
253
Eliminating s and τ from the first two results in (4.5.46) and substituting them into the first equation in (4.5.47) gives the solution
1 y(M 2 − 1) 1 1 x+ = x + y M2 − 1 2 . (4.5.48) u(x, y) = v M sin θ v The other choice of θ gives a negative q0 , and hence, the solution is given by 1 1 (4.5.49) u(x, y) = x − y M 2 − 1 2 . v Example 4.5.3 (The Schrödinger Equation in Quantum Mechanics). The striking feature of this example is that the optical wave equation (4.5.1) for the electromagnetic potential reduces to the Schrödinger equation in quantum mechanics. As indicated earlier, the eikonal function ψ is proportional to the characteristic function u. If u corresponds to ψ, then S = u − Et must be equal to a constant times the total phase of the optical wave which is given by the imaginary part of the exponent in (4.5.5) so that h (4.5.50) u − Et = k0 ψ(x) − c0 t , 2π h ) is constant. This shows that the total energy E is given by where ( 2π h h k0 c0 = ω = ω = hν, E= 2π 2π
(4.5.51)
where ω = 2πν and h = 2π. Equation (4.5.51) is the fundamental equation in quantum mechanics as h is the Planck constant. On the other hand, the wave speed c, frequency ν, and wavelength λ are related by λω c = λν = . (4.5.52) 2π According to the Hamilton–Jacobi theory, equation (4.4.54) relates c, E, and p. Consequently, (4.4.54) and (4.5.52) give another fundamental result in quantum mechanics: h h . (4.5.53) λ= = p 2m(E − V ) Thus, the optical wave equation for the potential φ with c = cn0 takes the form φtt − c2 ∇2 φ = 0.
(4.5.54)
If φ = ψ(x) exp(iωt) with (4.5.52), then (4.5.54) reduces to the form ∇2 ψ +
ω2 4π 2 2 ψ ≡ ∇ ψ + ψ = 0. c2 λ2
(4.5.55) 2
) by using (4.5.53) so Using (4.5.53) and (4.4.54), we can replace the factor ( 4π λ2 that (4.5.55) becomes 2m ∇2 ψ + 2 (E − V )ψ = 0. (4.5.56) This is the celebrated Schrödinger equation in quantum mechanics.
254
4 First-Order Nonlinear Equations and Their Applications
4.6 Exercises 1. Solve the following first-order partial differential equations: (a) pq = u with u(0, y) = y 2 , (b) p + q = pq with u(x, 0) = αx, where α is a constant, (c) xp2 + yq = u with u(s, 1) = −2s, where s is a parameter, (d) pq = xy with u(x, y) = −y at x = 0. 2. Obtain the complete integral of the following equations: (a) pq − u2 = 0, (b) p2 + q2 = u, 2 2 (d) p2 + qy − u = 0, (c) p + q + x = 0, (e) pq = xy, (f) x2 p2 + y 2 q 2 = u2 , 2 2 2 2 (h) yp2 (1 + x2 ) − qx2 = 0, (g) x p + y q = 4, 2 (i) 2q(u − px − qy) = 1 − q , (j) (p2 + q 2 )x − pu = 0. 3. Solve the following Cauchy problems: (a) p2 − qu = 0, u(x, y) = 1 on y = 1 − x, (b) px + qy − p2 q − u = 0, u(x, y) = x + 1 on y = 2, (c) 2p2 x + qy − u = 0, u(x, 1) = − 12 x, (d) 2pq − u = 0, u(x, y) = 12 y 2 at x = 0, (e) pq = 1, u(x, 0) = x, (f) pq = u, u(x, y) = 1 on y = −x, (g) u2x + uy = 0, u(x, 0) = x, (h) u − p2 + 3q 2 = 0, u(x, 0) = x2 , (i) cux + ut + acu = 0, u(x, 0) = f (x). 4. Find the solution of the equation in the parametric form xp + qy − p2 q − u = 0, with the initial data x(t, s) = s,
y(t, s) = 2,
u(t, s) = s + 1,
at t = 0.
5. Show that the complete integrals of the equation xpq + yq2 = 1, are (a) (u + b)2 = 4(ax + y) and (b) cx(u + d) = x2 + c2 y, where a, b, c, and d are constants. 6. Show that the integral surfaces of the nonlinear equation 2q(u − xp) − 2y(q + x) = 0, which are developable surfaces, are cones (u + ax)2 = 2y(x + b). 7. Show that the complete integrals of the equation 2xu + q 2 = x(xp + yq) are (a) u + a2 x = axy + bx2 and (b) x(y + cx)2 = 4(u − dx2 ), where a, b, c, and d are constants.
4.6 Exercises
255
8. Find the solution of the equation in the parametric form pq = xy,
u(x, y) = −y
when x = 0.
9. Show that no solution exists for the Cauchy problem pq = u,
u(x, y) = −1,
y = −x.
10. Show that the solution of the Cauchy problem u2x + ut = 0,
u(x, 0) = cx
is u(x, t) = cx − c2 t. 11. Find the solution of the following Cauchy problems: (a) xp + yq = xy, u(x, y) = 12 x2 on y = x. (b) −xp + yq = a (0 < x < y), u = 2x on y = 3x, where a is a constant. (c) (p2 + q2 )x = pu, u(0, s2 ) = 2s. (d) u2x + uy = 0, u(x, 0) = x. (e) 4ut − u2x = 4x2 , u(x, 0) = 0. 12. Obtain the solution of the equation 1 1 z p2 + q 2 2 − 1 − z 2 2 = 0
(p = zx andq = zy ),
so that the solution surface contains the line x(0, s) = s,
y(0, s) = sin θ,
z0 (s) = cos θ,
0<θ<
π . 2
13. Assuming u = u(x + ct) = u(ξ), show that the equation ∂u ∂u =0 , F u, ∂x ∂t reduces to the ordinary differential equation du du F u, = 0. ,c dξ dξ 14. Show that the complete solution of the equation z 2 p2 + q 2 + 1 = 1 (p = zx , q = zy ) is the family of cylinders of unit radius whose axis lies on the (x, y)-plane, that is, z 2 + (x cos θ + y sin θ + b)2 = 1, where a = tan θ and b is a constant of integration.
256
4 First-Order Nonlinear Equations and Their Applications
15. Solve the first-order nonlinear equations: (a) F (u, p, q, a) = (u − a)2 (p2 + q 2 + 1) − 1 = 0, (b) u2 (p2 + q 2 + 1) − 1 = 0, (c) p2 + q 2 = 1. 16. Solve the following Cauchy problems: (a) uux + uy = 0, u(x, 0) = x2 , (b) u2 ux + uy = 0, u(x, 0) = x. Show that the solution u(x, y) as y → 0 agrees with the initial condition in each case. 17. Show that the complete integral of the equation u2 p2 + q 2 + 1 = 1 is (x − a)2 + (y − b)2 + u2 = 1, where a and b are constants.
5 Conservation Laws and Shock Waves
The strides that have been made recently, in the theory of nonlinear partial differential equations, are as great as in the linear theory. Unlike the linear case, no wholesale liquidation of broad classes of problems has taken place; rather, it is steady progress on old fronts and on some new ones, the complete solution of some special problems, and the discovery of some brand new phenomena. The old tools—variational methods, fixed-point theorems, mapping degree, and other topological tools have been augmented by some new ones. Pre-eminent for discovering new phenomena is numerical experimentation; but it is likely that in the future numerical calculations will be parts of proofs. Peter Lax
5.1 Introduction Conservation laws describe the conservation of some basic physical quantities of a system and they arise in all branches of science and engineering. In this chapter, we study first-order, quasi-linear, partial differential equations which become conservation laws. We discuss the fundamental role of characteristics in the study of quasi-linear equations and then solve the nonlinear, initial-value problems with both continuous and discontinuous initial data. Special attention is given to discontinuous (or weak) solutions, development of shock waves, and breaking phenomena. As we have observed, quasi-linear equations arise from integral conservation laws which may be satisfied by functions which are not differentiable, and not even continuous, but simply bounded and measurable. These functions are called weak or generalized solutions, in contrast to classical solutions, which are smooth (differentiable) functions. It is shown that the integral conservation law can be used to derive the jump condition, which allows us to determine the speed of discontinuity or shock waves. Finally, a formal definition of a shock wave is given.
L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_5,
258
5 Conservation Laws and Shock Waves
Fig. 5.1 Volume V of a closed domain bounded by a surface S with surface element dS and outward normal vector n.
5.2 Conservation Laws A conservation law states that the rate of change of the total amount of material contained in a fixed domain of volume V is equal to the flux of that material across the closed bounding surface S of the domain. If we denote the density of the material by ρ(x, t) and the flux vector by q(x, t), then, the conservation law is given by d ρ dV = − (q · n) dS, (5.2.1) dt V S where dV is the volume element and dS is the surface element of the boundary surface S, n denotes the outward unit normal vector to S as shown in Figure 5.1, and the right-hand side measures the outward flux—hence, the minus sign is used. d inside the integral sign, Applying the Gauss divergence theorem and taking dt we obtain ∂ρ + div q dV = 0. (5.2.2) ∂t V This result is true for any arbitrary volume V , and, if the integrand is continuous, it must vanish everywhere in the domain. Thus, we obtain the differential form of the conservation law (5.2.3) ρt + div q = 0. In the presence of a source (or sink) function f (x, t, ρ), the total rate at which ρ is created (or destroyed) in the given domain is f (x, t, ρ) dV. (5.2.4) V
5.2 Conservation Laws
259
Inserting this term in (5.2.2), we obtain the integral form of the conservation law d ρ dV + divq dV = f (x, t, ρ) dV. (5.2.5) dt V V V An argument similar to that used before gives the differential form of the conservation law corresponding to (5.2.3) as ρt + div q = f (x, t, ρ),
(5.2.6)
where the flux term, div q, usually arises from the transport of ρ in the domain, and the source term f is often called a growth term in biological problems or a reaction term in chemistry. The one-dimensional version of the conservation law is ρt + qx = f (x, t, ρ).
(5.2.7)
From a mathematical or an empirical point of view, it is reasonable to assume a functional relation between q and ρ as a constitutive equation q = Q(ρ).
(5.2.8)
Thus, equation (5.2.7) and (5.2.8) form a closed system for ρ and q. Substituting (5.2.8) in (5.2.7) gives (5.2.9) ρt + c(ρ)ρx = f (x, t, ρ), where c(ρ) = dQ dρ . Equation (5.2.9) is universally considered the most fundamental, first-order, quasi-linear inhomogeneous wave equation. In particular, when f ≡ 0, (5.2.9) reduces to what is called the kinematic wave equation ρt + c(ρ)ρx = 0.
(5.2.10)
This equation often arises in nonlinear wave phenomena when the effects of dissipation, such as viscosity and diffusion, are neglected. We next investigate the development of shocks from the initial-value problem for u(x, t) ut + c(u)ux = 0, −∞ < x < ∞, t > 0, u(x, 0) = f (x), −∞ < x < ∞,
(5.2.11) (5.2.12)
where c(u) and f (x) are C 1 (R) functions of their arguments, that is, they are smooth functions. The characteristic equations associated with (5.2.11) are dx du dt = = . 1 c(u) 0 These equations give
260
5 Conservation Laws and Shock Waves
Fig. 5.2 Characteristic lines of different slopes.
du = 0 and dt dx = c(u). dt
(5.2.13a) (5.2.13b)
Clearly, the solution of (5.2.13b) represents characteristics of equation (5.2.11). Along these characteristics, du dx = ux + ut = c(u)ux + ut = 0. dt dt This means that u is constant on the characteristics which propagate with speed c(u). The dependence of c on u produces a gradual nonlinear distortion of the wave profile as it propagates in the medium. It also follows that c(u) is constant on the characteristics, and therefore, the characteristics must be straight lines in the (x, t)-plane with constant slope 1/c(u). Equations of these lines are given by x − tc(u) = const. = A,
(5.2.14)
where A is a constant, that is, x = x(t) = tc(u) + A. If any one of these characteristics intersects that x-axis (t = 0) at x(0) = ξ, then, by the initial condition, u(ξ, 0) = f (ξ). Thus, the equation of a typical characteristic line (see Figure 5.2) joining two points (ξ, 0) and (x, t) is x = ξ + tF (ξ),
(5.2.15)
where F (ξ) = c(f (ξ)). Since u(x, t) is constant on the characteristics, it follows from (5.2.15) and the initial condition that u(x, t) = u ξ + tF (ξ), t = u(ξ, 0) = f (ξ). Thus, if a solution of the initial-value problem exists for t > 0, then the solution can be written in the parametric form
5.2 Conservation Laws
u(x, t) = f (ξ), ξ = x − tF (ξ),
261
(5.2.16ab)
where F (ξ) = c(f (ξ)). If there are two points (ξ, 0) and (η, 0), with ξ < η and m1 =
1 1 < = m2 , F (ξ) F (η)
then the characteristics starting at (ξ, 0) and (η, 0) will intersect at the point P (x, t) for t > 0. At the point of intersection P (x, t), the solution u(x, t) has two different values f (ξ) and f (η). This means that u is double valued, and hence, the solution is not unique at the point of intersection of the characteristics. Thus, the solution must be discontinuous at the point of intersection. The conclusion is that if no two characteristic lines intersect in the half plane t > 0, there exists a solution of the initial-value problem (5.2.11), (5.2.12) as a differentiable function for all t > 0. This can happen only if the reciprocal of the slope is an increasing function of the intercept, that is, F (ξ) ≤ F (η) for ξ ≤ η. (5.2.17) In other words, the family of characteristics spreads only for t > 0 and generates a solution of the problem that is at least as smooth as f (x). Such a solution is called an expansive (or refractive) wave. We now verify that (5.2.16ab) represents an analytical solution of the problem. Differentiating (5.2.16ab) with respect to x and t, we obtain ux = f (ξ)ξx , ut = f (ξ)ξt , 1 = {1 + tF (ξ)}ξx , 0 = F (ξ) + {1 + tF (ξ)}ξt . Eliminating ξx and ξt gives ux =
f (ξ) , 1 + tF (ξ)
ut = −
F (ξ)f (ξ) . 1 + tF (ξ)
(5.2.18ab)
Substituting ux and ut , equation (5.2.11) is satisfied provided {1 + tF (ξ)} = 0. The solution (5.2.16ab) also satisfies the initial condition at t = 0 since ξ = x, and hence, it is unique. Suppose that u(x, t) and v(x, t) are two solutions. Then, on x = ξ + tF (ξ), u(x, t) = u(ξ, 0) = f (ξ) = v(x, t). Thus, we proved the following. Theorem 5.2.1. The nonlinear initial-value problem given by (5.2.11), (5.2.12) has a unique solution provided that {1 + tF (ξ)} = 0 and f and c are C 1 (R) functions where F (ξ) = c(f (ξ)). The solution is given by the parametric form (5.2.16ab). Remark. When c(u) = const. = c > 0, we obtain the linear initial-value problem ut + cux = 0,
x ∈ R, t > 0,
(5.2.19)
262
5 Conservation Laws and Shock Waves
Fig. 5.3 Parallel characteristic lines.
u(x, 0) = f (x),
x ∈ R.
(5.2.20)
This linear problem has a unique solution given by u(x, t) = f (x − ct).
(5.2.21)
This solution represents a traveling wave moving with constant velocity c in the positive direction of the x-axis without any change of shape. In this case, the characteristics x = ξ + ct form a family of parallel straight lines in the (x, t)-plane as shown in Figure 5.3. •
Physical Significance of Solution (5.2.16ab).
For the general nonlinear problem, the dependence of the wave speed c on u produces a gradual nonlinear distortion of the wave as it propagates in the medium. This means that some parts of the wave travel faster than others. When c (u) > 0, c(u) is an increasing function of u. In this case, higher values of u propagate faster than lower ones. On the other hand, when c (u) < 0, c(u) is a decreasing function of u, and higher values of u travel slower than the lower ones. This means that the wave profile progressively distorts itself, leading to a multi-valued solution with a vertical slope, and hence, it breaks. In the linear case, c is constant, there is no such distortion of the wave, and hence, it propagates without any change of shape. Thus, there is a striking difference between the linear and nonlinear solutions. At any compressive part of the wave, the wave speed is a decreasing function of x, as shown in Figure 5.4. The wave profile distorts progressively to produce a triple-valued solution for u(x, t), and hence, it ultimately breaks. It follows from Theorem 5.2.1 that the solution of the nonlinear initial-value problem exists provided that 1 + tF (ξ) = 0. This condition is always satisfied for a sufficiently small time t. It also follows from (5.2.18ab) that both ux and ut tend to infinity as 1 + tF (ξ) → 0. This means that the solution develops a discontinuity (singularity) when 1 + tF (ξ) = 0. Thus, on any characteristic for which F (ξ) < 0, a discontinuity occurs at time t given by
5.2 Conservation Laws
263
Fig. 5.4 Distortion of wave profile with increasing time t2 t1 > 0.
t=−
1 , F (ξ)
(5.2.22)
which is positive because F (ξ) = c (f )f (ξ) < 0. If we assume c (f ) > 0, this inequality implies that f (ξ) < 0. Hence, the solution (5.2.16ab) ceases to exist for all time if the initial data is such that f (ξ) < 0 for some value of ξ. Suppose that t = τ is the time when the solution first develops a discontinuity (singularity) for some value of ξ. Then, τ =−
1 > 0. min−∞<ξ<∞ {c (f )f (ξ)}
(5.2.23)
Thus, the shape of the initial curve for u(x, t) changes continuously with increasing values of t, and the solution becomes multi-valued with a vertical slope for t ≥ τ . Therefore, the solution breaks down when F (ξ) < 0 for some ξ, and such breaking is a strikingly nonlinear phenomenon. Indeed, Whitham (1974) emphasized that: “This breaking phenomenon is one of the most intriguing long-standing problems of water wave theory.” In the linear theory, such breaking will never occur. More precisely, the development of a discontinuity in the solution for t ≥ τ can also be seen in the (x, t)-plane. If f (ξ) < 0, then we can find two values of ξ = ξ1 , ξ2 (ξ1 < ξ2 ) on the initial line (t = 0) such that the characteristics through them have different slopes 1/c(u1 ) and 1/c(u2 ), where u1 = f (ξ1 ), u2 = f (ξ2 ), and c(u2 ) > c(u1 ). These two characteristics will intersect at a point in the (x, t)-plane for some t > 0. Since the characteristics carry constant values of u, the solution ceases to be single-valued at the point of intersection. As Figure 5.4 shows, the solution u(x, t) progressively distorts itself, and, at any instant of time, there exists an interval on the x-axis where u becomes triple-valued for a given x. The end result is the development of multi-valued solutions, and hence, it leads to breaking. This is exactly the situation always observed on beaches when water waves break. Finally, we conclude the above discussion by stating the remarkable fact that both the distortion of the wave profile and the development of a discontinuity or a shock are typical nonlinear phenomena. Therefore, when 1 + tF (ξ) = 0, the solution develops a discontinuity known as a shock. The analysis of a shock involves an extension of a solution to allow for
264
5 Conservation Laws and Shock Waves
discontinuities. It is also necessary to impose certain restrictions on the solution to be satisfied across its discontinuity. Example 5.2.1. Solve the nonlinear initial-value problem ut + uux = 0, u(x, 0) =
x ∈ R, t > 0, (a − x ) if |x| ≤ a, 0 if |x| ≥ a. 2
(5.2.24)
2
(5.2.25)
In this case, c(u) = u, and the solution follows from (5.2.16ab) as (a2 − ξ 2 ) if |x| ≤ a, u(x, t) = 0 if |x| ≥ a, where
ξ = x − t a2 − ξ 2 .
This is a quadratic equation in ξ giving
1 1 1 ± 1 − 4t x − ta2 2 , t = 0. ξ= 2t The solution of (5.2.24) becomes, for |ξ| ≤ a, u(x, t) = a2 − ξ 2 ,
1 1 = 2 2xt − 1 ± 1 − 4xt + 4a2 t2 2 , t = 0, 2t and, for |ξ| ≥ a, u(x, t) = 0.
(5.2.26)
For small values of t (t → 0), only the positive sign before the radical in (5.2.26) is acceptable so that the initial condition is satisfied. On the other hand, when t > T , both signs are admissible for x > a. We next draw the characteristics in the (x, t)-plane for a = 1 in Figure 5.5. As stated before, characteristics are straight lines with speed u. Several characteristic lines intersect. At the point where two characteristics intersect, the solution becomes double-valued, and the slope in the (x, t)-plane becomes infinite. From this point onward, the solution is a discontinuous function of position, and it corresponds to the onset of a shock wave. Moreover, it follows that all the characteristics, originating from (x, 0) where x > −1, intersect the characteristics starting from the point x, where x ≥ 1, at some point or another. In particular, the two characteristics initially at x = 0 and x = 1 intersect at the point (x, t) = (1, 1). The solution would be double-valued at (1, 1). Figure 5.6 represents the propagation of the initial parabolic pulse with a = 1. As t increases, the initial pulse distorts progressively. This progressive change in the initial wave pulse is the result of the nonlinear term in the equation. In the linear case (ut + cux = 0, c = const.) with the same initial data (5.2.25) for a = 1, the initial parabolic pulse propagates with constant velocity c in the positive direction of the x-axis without change of shape.
5.2 Conservation Laws
Fig. 5.5 Characteristics of (5.2.24) with the condition for a = 1.
Fig. 5.6 Propagation of a parabolic pulse for a = 1.
Example 5.2.2. Solve the initial-value problem ut + uux = 0, x ∈ R, t > 0, u(x, 0) = x2 , x ∈ R. According to (5.2.16ab), the solution is given by u(x, t) = f (ξ) = ξ 2 ,
ξ = x − tξ 2 .
Hence, the second equation becomes tξ 2 + ξ − x = 0. Solving this quadratic equation in ξ, we obtain
265
266
5 Conservation Laws and Shock Waves
ξ=
√ 1 −1 ± 1 + 4xt 2t
for t = 0.
(5.2.27)
Thus, the solution is
√ 2 1 −1 ± 1 + 4xt 2 4t √ 1 = 2 1 + 2xt ± 1 + 4xt , t = 0. 2t
u(x, t) = ξ 2 =
(5.2.28)
This solution must satisfy the initial data x2 = u(x, 0) = lim u(x, t). t→0
With the positive sign before the radical in the solution (5.2.28), this limit does not exist. With the negative sign before the radical, the limit exists by using the L’Hospital rule twice with fixed x. Thus, the solution is u(x, t) =
√ 1 1 + 2xt − 1 + 4xt , 2 2t
t = 0.
(5.2.29)
This represents a solution only in the region 1+4xt ≥ 0, which is the region between two branches of the hyperbola xt = − 14 in the (x, t)-plane. Example 5.2.3. Solve the initial-value problem ut + uux = 1, u(s, 2s) = s,
x ∈ R, t > 0, s is a parameter.
(5.2.30) (5.2.31)
The characteristic equations are dx du dt = = = dτ. 1 u 1
(5.2.32)
The initial data at τ = 0 are x(0, s) = s,
t(0, s) = 2s,
u(0, s) = s.
Thus, the solutions of this system (5.2.32), (5.2.33) are ⎫ 2 x(τ, s) = τ2 + sτ + s, ⎪ ⎬ t(τ, s) = τ + 2s, ⎪ ⎭ u(τ, s) = τ + s.
(5.2.33)
(5.2.34)
Eliminating τ from the first two results gives the characteristics in the (x, t)-plane as (t − s)2 = 2(x − s) + s2 ,
(5.2.35)
where s is a parameter. This represents a family of parabolas for s = 0, ±1, ±2, . . . . We leave the construction of the diagram for parabolas as an exercise for the reader.
5.2 Conservation Laws
267
Fig. 5.7 Compression wave with overlapping characteristics.
We first use (5.2.35) to find s and then use the second result of (5.2.34) to find τ and s in terms of x and t. This gives (t2 − 2x) 2x − t , s= . (5.2.36) τ= t−1 2(t − 1) Substituting these results into the third equation in (5.2.34) gives the integral surface of the equation (5.2.30) as u(x, t) =
(2x − 2t + t2 ) . 2(t − 1)
(5.2.37)
Thus, the initial-value problem has the solution (5.2.37) everywhere in the (x, t)plane except on the straight line t = 1. Example 5.2.4 (Initial-Value Problem with Discontinuous Initial Data). We consider the initial-value problem for equation (5.2.11) with discontinuous initial data at t = 0, u1 if x < 0, (5.2.38) u(x, 0) = u2 if x > 0,
and F (x) =
c1 = c(u1 ) c2 = c(u2 )
if x < 0, if x > 0,
(5.2.39)
where u1 and u2 are constants. There are two cases: (i) u1 > u2 and (ii) u1 < u2 . Case (i) u1 > u2 with c1 > c2 . In this case, breaking will occur almost immediately, and this can be seen from Figure 5.7. The multi-valued region starts right at the origin and is bounded by the characteristics x = c2 t and x = c1 t. This corresponds to a centered compression wave with overlapping characteristics in the (x, t)-plane.
268
5 Conservation Laws and Shock Waves
Fig. 5.8 Centered expansive wave.
Case (ii) u1 < u2 with c1 < c2 . In this case, the initial condition is expansive with c1 < c2 . A continuous solution can be found from (5.2.16ab) in which all values of F (x) in [c1 , c2 ] are taken on the characteristics through the origin ξ = 0. This corresponds to a centered fan of characteristics x = ct where c1 ≤ c ≤ c2 in the (x, t)-plane so that the solution has the explicit form (see Figure 5.8) c=
x , t
c1 <
x < c2 . t
Thus, the complete solution is given by ⎧ ⎪ ⎨ c1 if x ≤ c1 t, c = xt if c1 t < x < c2 t, ⎪ ⎩ c2 if x ≥ c2 t.
(5.2.40)
(5.2.41)
Example 5.2.5. Solve the initial-value problem with discontinuous initial data ut + uux = 0, x ∈ R, t ≥ 0; ⎧ if x ≤ 0, ⎪ ⎨1 u(x, 0) = 1 − x if 0 ≤ x ≤ 1, ⎪ ⎩ 0 if x ≥ 1. According to the parametric solution (5.2.16ab), we obtain the solution ⎧ if ξ ≤ 0, ⎪ ⎨1 u(x, t) = f (ξ) = 1 − ξ if 0 ≤ ξ ≤ 1, ⎪ ⎩ 0 if ξ ≥ 1, ⎧ if ξ ≤ 0, ⎪ ⎨ξ + t u(ξ, t) = ξ + tf (ξ) = ξ(1 − t) + t if 0 ≤ ξ < 1, ⎪ ⎩ ξ if ξ ≥ 1.
(5.2.42) (5.2.43)
(5.2.44)
(5.2.45)
5.2 Conservation Laws
269
Fig. 5.9 Characteristics and graphical representation of solution.
Fig. 5.10 Weak or generalized solution.
The solution and the characteristics are drawn in various regions, as shown in Figure 5.9. In 0 ≤ ξ ≤ 1, the solution is u(x, t) = 1−x 1−t . This is not defined at t = 1. The characteristics are straight lines intersecting the x-axis at (ξ, 0), and u is constant on a given characteristic line. The values of u on t = 0 are propagated along the characteristics. As shown in Figure 5.9, the characteristics are drawn in various regions in the (x, t)-plane. The value of u on a characteristic line originating from a point (ξ, 0) is 1 − ξ, where 0 ≤ ξ ≤ 1, and these characteristics all intersect at (1, 1). This means that u is multi-valued at (1, 1), and hence, the solution breaks down at this point. It also follows from Figure 5.9 that characteristics originating from (ξ, 0), with ξ < 0, intersect those originating from (ξ, 0), with ξ > 1. As u has different values on these characteristics, it follows that it is not defined in the quadrant x ≥ 1, t ≥ 1. Figure 5.9 also suggests that the most likely weak solution is one which is discontinuous across some curve originating from (1, 1) with u = 1 to its left and u = 0 to its right. In conservation form, the given equation can be written as 1 2 u = 0, (5.2.46) ut + 2 x where the flux is 12 u2 . It follows from this equation combined with the shock condition that the slope of the shock path is 2. Thus the equation of the shock path is t = 2x − 1, as shown in Figure 5.10.
270
5 Conservation Laws and Shock Waves
5.3 Discontinuous Solutions and Shock Waves It should be pointed out that the nonlinear conservation equation ρt + qx = 0
(5.3.1)
has been solved under two basic assumptions: (i) there exists a functional relation q = Q(ρ), and (ii) ρ and q are continuously differentiable. When breaking occurs, questions arise about the validity of these assumptions. To examine the development of discontinuities or shocks, we still assume that q = Q(ρ), but will allow jump discontinuity at x = s(t). We also assume x1 and x2 such that x1 < s(t) < x2 and ds dt = U (t). Without any source (or sink), we assume that the integral form of the conservation equation (5.2.5) holds for the one-dimensional case and has the form d x2 ρ dx = q(x1 , t) − q(x2 , t), (5.3.2) dt x1 or d dt
s−
x2
ρ dx +
ρ dx + q(x2 , t) − q(x1 , t) = 0.
(5.3.3)
s+
x1
This implies that
s−
x1
ρt dx + sρ ˙ s− , t +
x2
ρt dx − sρ ˙ s+ , t + q(x2 , t) − q(x1 , t) = 0, (5.3.4)
s+
where ρ(s− , t), ρ(s+ , t) are the values of ρ(x, t), as x → s from below and above, respectively. Since ρt is bounded on x1 < x < s− and s+ < x < x2 , respectively, the integrals in (5.3.4) must vanish as x1 → s− and x2 → s+ . Consequently, in the limit,
(5.3.5) q s− , t − q s+ , t = U ρ s− , t − ρ s+ , t . In terms of the conventional notation of shock dynamics, condition (5.3.5) can be written as [q] = U [ρ], (5.3.6) where [q] = (q − − q+ ) and [ρ] = (ρ− − ρ+ ) denote the jump in q and ρ, respectively, across the discontinuity x = s(t), as shown in Figure 5.11. Result (5.3.6) is usually called the jump (shock) condition, which gives a relation ahead (right) of and behind (left) of the discontinuity and the speed of the discontinuity U (t) (reciprocal of the slope). The discontinuity in ρ that propagates along the curve x = s(t) is known as the shock wave, and x = s(t) is simply called the shock. In gas dynamics, (5.3.6) is known as the Rankine–Hugoniot condition for the speed of the shock wave. Thus, the basic problem can be written as dρ dq + =0 dt dx
at points of continuity,
(5.3.7)
5.4 Weak or Generalized Solutions
271
Fig. 5.11 Discontinuity at x = s(t).
−U [ρ] + [q] = 0
at points of discontinuity.
(5.3.8)
This leads to a nice correspondence: d ↔ −U [ ], dt
d ↔[] dx
(5.3.9ab)
between the differential equation and the shock condition. We can now generalize the concept of the solution for equation (5.3.1) to include discontinuous solutions. This kind of extension of a solution is not a purely mathematical exercise, as the notion of a shock is of great significance in applied areas, such as fluid dynamics, gas dynamics, and plasma physics. From a physical point of view, a shock represents an idealization of a thin transition region separating two regions where solutions of the basic flow equations can readily be determined. Now, it is possible to find discontinuous solutions of equation (5.3.1). In any continuous part of the solution, equation (5.3.1) is still satisfied, and the assumption q = Q(ρ) can be retained. Thus, q − = Q(ρ− ) and q + = Q(ρ+ ) hold on the two sides of a shock, and the shock condition (5.3.6) can be written as (5.3.10) U ρ− − ρ+ = Q ρ− − Q ρ+ . This leads to a problem of fitting shock discontinuities into the solution (5.2.16ab) so that (5.3.10) is satisfied and multi-valued solutions can be avoided.
5.4 Weak or Generalized Solutions We formally examine the possibility of generalizing the concept of a classical solution so as to include discontinuous and nondifferentiable functions. We now consider the general procedure necessary to define what is usually called a weak solution of the nonlinear, initial-value problem ρt + qx = 0, x ∈ R, t > 0, ρ(x, 0) = f (x), x ∈ R, where q(x, t) is a continuously differentiable function on R.
(5.4.1) (5.4.2)
272
5 Conservation Laws and Shock Waves
Fig. 5.12 Compression wave with overlapping characteristics.
A classical solution of the partial differential equation (5.4.1) is a smooth function ρ = ρ(x, t) which satisfies (5.4.1). We assume that ρ(x, t) is a classical solution of (5.4.1). We consider a class of test functions φ = φ(x, t) such that φ ∈ C ∞ and it has a compact support in the (x, t)-plane. We can choose an arbitrary rectangle D = {(x, t) : a ≤ x ≤ b, 0 ≤ t ≤ T } where φ = 0 outside D, and on the boundary lines x = a, x = b, and t = T , as shown in Figure 5.12. Equation (5.4.1) is multiplied by φ(x, t) and integrated over D to obtain (ρt + qx )φ dx dt = 0. (5.4.3) D
Integrating both terms by parts gives b T T [φρ]0 − ρφt dt dx + a
or
0
b
φ(x, 0)f (x) dx −
− a
[φq]ba
b
−
0
b
T
qφx dx dt = 0,
a
T
b
T
ρφt dt dx − a
0
qφx dt dx = 0. a
0
Thus the final form of the above equation is (ρφt + qφx ) dx dt + φ(x, 0)f (x) dx = 0. DU t≥0
(5.4.4)
R
This holds for all test functions φ(x, t) in the half plane. Result (5.4.4) does not involve any derivatives of ρ or q. Indeed, it remains valid even if ρ and q or their derivatives are discontinuous. Thus, we have proved that, if ρ(x, t) is a classical solution of the problem (5.4.1), (5.4.2), then (5.4.4) holds for all test functions φ with compact support in the half (x, t)-plane. The functions ρ(x, t), which satisfy (5.4.4) for all test functions φ, are called weak or generalized solutions of the problem. This leads to the following. Definition 5.4.1 (Weak Solution). A bounded measurable function ρ(x, t) is called a weak solution of the initial-value problem (5.4.1), (5.4.2) with bounded and measurable initial data f (x), provided that (5.4.4) holds for all test functions φ(x, t) with compact support in the half plane.
5.4 Weak or Generalized Solutions
273
This is a more general definition of a solution that does not require the smoothness property. To show that the concept of a weak solution is, indeed, a generalization of a classical solution, we prove the following. Theorem 5.4.1. If (5.4.4) holds for all test functions φ with compact support for t > 0 and if u is smooth, then ρ is a classical solution of the initial-value problems (5.4.1) and (5.4.2). Proof. Since φ has compact support for t > 0, then we integrate (5.4.4) by parts to obtain (ρt + qx )φ dx dt = 0. (5.4.5) RU t>0
This is true for all test functions φ, and hence, ρt + qx = 0 for x ∈ R, t > 0.
(5.4.6)
We next multiply (5.4.6) by φ and integrate the resulting equation by parts over the region D to find (ρφt + qφx ) dx dt + φ(x, 0)ρ(x, 0) dx = 0. (5.4.7) RU t>0
R
Subtracting (5.4.7) from (5.4.4) gives ρ(x, 0) − f (x) φ(x, 0) dx = 0.
(5.4.8)
R
This is true for all test functions φ(x, 0). Since f (x) is continuous, (5.4.8) leads to ρ(x, 0) = f (x). This shows that ρ(x, t) is the classical solution of the initial-value problem. We next consider only those weak solutions ρ(x, t) which are continuously differentiable in two parts D1 and D2 of a domain D, but with a jump discontinuity (shock) along the dividing smooth curve Γ between D1 and D2 , as shown in Figure 5.13. Suppose that the equation of Γ is x = s(t). Since ρ(x, t) has a jump discontinuity along this curve, x = s(t), ρ(x, t) is smooth away from Γ , and limits ρ− (t) = ρ(s−, t) and ρ+ (t) = ρ(s+, t) from the left and the right exist. The speed of the shock wave is s(t), ˙ which is the reciprocal of the slope in Figure 5.13. For any test function φ, result (5.4.4) becomes (ρφt + qφx ) dx dt = (ρφt + qφx ) dx dt 0= D D1 (ρφt + qφx ) dx dt. (5.4.9) + D2
Since ρ(x, t) is smooth in D1 and in D2 , for n = 1, 2, the divergence theorem gives
274
5 Conservation Laws and Shock Waves
Fig. 5.13 Smooth curve Γ across which ρ(x, t) has a jump discontinuity.
(ρφt + qφx ) dx dt = Dn
(ρφ)t + (qφ)x dx dt
Dn
=
φ(−ρ dx + q dt). ∂Dn
Because φ = 0 on the boundary ∂D, these line integrals vanish everywhere, except on Γ . Thus, p2 φ(−ρ dx + q dt) = φ −ρ− dx + q − dt D p1 p2 1 φ(−ρ dx + q dt) = − φ −ρ+ dx + q + dt . D2
p1
Therefore, equation (5.4.9) gives φ −[ρ] dx + [q] dt = 0.
(5.4.10)
Γ
This result is true for all test functions φ, and therefore, the integrand must vanish, that is, on each point on Γ , s[ρ] ˙ = [q]. (5.4.11) This is the same jump condition (5.3.6). Thus, a weak solution leads to the jump condition across a jump discontinuity. Example 5.4.1. Solve the nonlinear Cauchy problem ut + uux = 0,
x ∈ R, t > 0,
with two different sets of discontinuous initial data 1 if x < 0, (i) u(x, 0) = 0 if x > 0, 0 if x < 0, (ii) u(x, 0) = 1 if x > 0.
(5.4.12)
5.4 Weak or Generalized Solutions
275
Fig. 5.14 Initial data, characteristics, and a discontinuous solution.
The conservation law associated with equation (5.4.12) is 2 u = 0. ut + 2 x
(5.4.13)
In case (i), c(u) = u, c(u(x, 0)) is a decreasing function of x, and condition (5.2.17) is violated. Thus, there is no continuous solution of the initial-value problem. However, there exists a discontinuous solution given by 0 if x < U t, (5.4.14) u(x, t) = 1 if x > U t, which satisfies (5.4.12) for x = U t and (5.4.11) on the line x = U t. Hence, (5.4.14) is called a generalized or weak solution. This solution has a jump discontinuity along the line x = U t, where U is the speed of the shock wave given by U=
( 12 · 02 − 12 · 12 ) (q− − q + ) 1 = = . (u− − u+ ) (0 − 1) 2
The initial data, characteristics, and the weak solution are shown in Figure 5.14. In case (ii), c(u) = u, c(u(x, 0)) is an increasing function of x, and condition (5.2.17) is satisfied. Hence, there is a continuous solution of (5.4.12) (except at x = t = 0) with the initial data (ii) given by ⎧ ⎪ ⎨ 0 if x < 0, (5.4.15) u(x, t) = xt if 0 < x < t, ⎪ ⎩ 1 if x > t. A solution of the form (5.4.15) is called an expansive or refractive wave. Often this solution in the sector 0 < x < t is referred to as a centered simple wave. The boundary lines x = 0 and x = t of the centered simple wave are weak waves which propagate along the characteristics, as shown in Figure 5.15. This represents a solution of (5.4.12) for t ≥ 0 which can be verified by direct substitution, x x x x x + = − 2 + 2 = 0. t t t t x t t
276
5 Conservation Laws and Shock Waves
Fig. 5.15 Initial data, characteristics, and the continuous solution.
Fig. 5.16 Characteristics and the discontinuous solution.
The initial data, characteristics, and the solution are shown in Figure 5.15. On the other hand, a remarkable fact involved in this nonlinear Cauchy problem with discontinuous data (ii) is that there is also a discontinuous solution given by 0 if 2x < t, u(x, t) = (5.4.16) 1 if 2x > t. Obviously, this represents a solution everywhere in the (x, t)-plane, except on the line of discontinuity t = 2x. The jump condition gives the shock speed U=
( 12 · 02 − 12 · 12 ) 1 = . (0 − 1) 2
The characteristics and discontinuous solution are drawn in Figure 5.16. Thus, there are at least two solutions of equation (5.4.12) with the initial data (ii), and hence, the solution is not unique. The main question is: Which one of these solutions represents a physically meaningful solution? Obviously, another mathematical criterion is required for determining a unique weak solution. It seems that the criterion for determining a physically meaningful weak solution is closely related to the search for a criterion for an acceptable discontinuous solution. It is also clear from
5.5 Exercises
277
the above analysis that, in general, a discontinuous solution arises whenever characteristics of a nonlinear equation intersect in the (x, t)-plane. It can be proved that there is a unique weak discontinuous solution of the Cauchy problem which satisfies the following inequality on the curve of discontinuity: (5.4.17) a u− > U > a u+ . This is called the Lax entropy criterion. Mathematically, this criterion means that the wave speed just behind the shock is greater than the wave speed just ahead of it. In other words, the wave behind the shock catches up to the wave ahead of it. This entropy criterion is a special case of the second law of thermodynamics: entropy increases across a shock. Geometrically, the criterion for a unique weak discontinuous solution can be stated as follows: The characteristics originating on either side of the discontinuity curve, when continued in the direction of increasing t, intersect the curve of discontinuity. Clearly, solution (5.4.14) satisfies the entropy inequality (5.4.17), whereas solution (5.4.16) does not satisfy (5.4.17). Therefore, the former is an acceptable discontinuous solution, but the latter is not. This leads to a formal definition of a shock wave. A discontinuity which satisfies both the jump condition (5.4.11) and the inequality (5.4.17) on its curve of discontinuity is called a shock. The main problem is to investigate whether every nonlinear initial-value problem has a unique weak solution defined for all t ≥ 0 with only shock as a discontinuity. The proof of the existence and uniqueness of a weak solution is fairly difficult and is beyond the scope of this book, but is provided by Lax (1973).
5.5 Exercises 1. Find the solution of the initial-value problems (a) ut + uux = 0, x ∈ R, t > 0, u(x, 0) = sin x, x ∈ R, (b) ut + uux = 0, x ∈ R, t > 0, u(x, 0) = −x, x ∈ R. 2. Show that the equation ut + uux = 0 gives an infinite number of conservation laws ∂ ∂ n n u + un+1 = 0, ∂t ∂x n + 1 where n = 1, 2, 3, . . . . Hence, deduce the integral conservation law x2 n+1 n ∂ un dx + (x1 ) − un+1 (x2 ) = 0, u ∂t x1 n+1 where x1 and x2 are two fixed points.
278
5 Conservation Laws and Shock Waves
3. Show that the solution of the initial-value problem ut + uux = 0,
x ∈ R, t > 0,
u(x, 0) = bx,
x ∈ R,
where a and b are constants, is u(x, t) =
abx exp(−at) . (a + b) − b exp(−at)
4. Find the solution of the initial-value problem ut + uux = 0, u(x, 0) =
x ∈ R, t > 0, 0 exp(− x1 )
if x ≤ 0, if x > 0.
5. Solve the initial-value problem ut + cux = 0, x ∈ R, t > 0, −1 , x ∈ R, u(x, 0) = a2 + x2 where c and a are constants. 6. Show that the solution of the Cauchy problem ut − 2axtux = 0, x ∈ R, t > 0, u(x, 0) = f (x), x ∈ R, is
u(x, t) = f x exp at2 .
Draw the characteristics x = ξ exp(−at2 ), where ξ is a constant. 7. Show that the equation ut + uux = 1,
x ∈ R, t > 0,
has no solution such that u = 12 t, when t2 = 4x, and has no unique solution, such that u = t when t2 = 2x. 8. Find a weak solution of the initial-value problem in t ≥ 0 ut + uux = 0, u(x, 0) =
x ∈ R, t > 0, 0
if x < 0 or x > 1,
1
if 0 ≤ x ≤ 1.
9. Find the solution of the initial-value problem in t > 0 ut + uux = 0, x ∈ R, t > 0, ⎧ if x ≤ 0, ⎪ ⎨ −1 u(x, 0) = 2x − 1 if 0 ≤ x ≤ 1, ⎪ ⎩ 1 if x ≥ 0.
5.5 Exercises
10. Solve the initial-value problem in t > 0 ut + uux = 0, x ∈ R, t > 0, ⎧ if x ≤ 0, ⎪ ⎨1 u(x, 0) = 1 − 2x if 0 ≤ x ≤ 1, ⎪ ⎩ −1 if x ≥ 0. 11. Find the weak solution for t ≥ 0 of the initial-value problem ut + uux = 0, u(x, 0) =
x ∈ R, t > 0, −a 2a
if x ≤ 0, if x > 0,
where a < 0 and a > 0. 12. Show that the solution of the initial-value problem ut + u2 ux = 0,
is u(x, t) =
x 1 ( 2t
x ∈ R, t > 0,
u(x, 0) = x, if t = 0,
√ 1 + 4xt − 1)
if t = 0 and 1 + 4xt > 0.
13. Solve the equation ut + uux = 0, for two cases:
x ∈ R, t > 0,
⎧ if x ≤ 0, ⎪ ⎨1 (a) u(x, 0) = (1 − xa ) if 0 < x < a, ⎪ ⎩ 0 if x ≥ a; ⎧ ⎪ ⎨ 0 if x ≤ 0, (b) u(x, 0) = xa if 0 < x < a, ⎪ ⎩ 1 if x ≥ a.
Examine both cases in the limit, as a → 0. 14. Solve the initial-value problem ut + uux = 0, u(x, 0) =
x ∈ R,
x ∈ R, t > 0, 1 if |x| < 1, 0 if |x| > 1.
15. Solve the initial-value problem for u = u(x, t) ut + u2 ux = 0, x ∈ R, t > 0, √ u(x, 0) = x, x ∈ R.
279
280
5 Conservation Laws and Shock Waves
16. Show that the solution of the initial-value problem ut + u3 ux = 0, x ∈ R, t > 0, 1 u(x, 0) = x 3 , x ∈ R, is
u(x, t) =
x 1+t
13 .
Examine the solution as t → ∞. 17. Show that the solution of the initial-value problem ut + uux = 2t, x ∈ R, t > 0, u(x, 0) = x, x ∈ R, is u(x, t) = t2 +
3x − t3 . 3(1 + t)
18. Show that the solution of the initial-value problem ut + 3tux = u, x ∈ R, t > 0, u(x, 0) = cos x, x ∈ R, 3 u(x, t) = et cos x − t2 . 2
is
19. Show that the solution of the initial-value problem 2ut + ux = 0, u(x, 0) =
is u(x, t) =
x ∈ R, t > 0, sin x
if 0 ≤ x ≤ π,
0
otherwise
sin(x − 12 t) 0
if 12 t ≤ x ≤ 12 t + π, otherwise.
20. Use the method of characteristics to solve the initial boundary-value problem ut + ux = x, x > 0, t > 0, u(x, 0) = sin x, x > 0, u(0, t) = t, t > 0. Show that the solution is u(x, t) =
1 2 2x
− (x − t)
t(x −
1 2 t)
+ sin(x − t)
if x ≤ t, if x > t.
5.5 Exercises
281
21. Show that the implicit solution of the equation x2 ux + uut = 1, u = 0 on x + t = 1, x > 0, is
1 2 u + 1 − t (1 + ux) = x. 2
22. The conservation form of the Buckley–Leverett (1942) equation for saturation, s, for propagation of non-aqueous phase liquids (typically, hydrocarbons) and water in porous media is st + Fx = 0,
or
st + csx = 0,
where F = Vd s2 /[s2 + b(1 − s)2 ], Vd is the Darcy velocity, c = ∂F ∂s is the propagation speed, and b is the shape parameter. Show that s is conserved. Explain the features of the speed c.
6 Kinematic Waves and Real-World Nonlinear Problems
. . . as Sir Cyril Hinshelwood has observed . . . fluid dynamicists were divided into hydraulic engineers who observed things that could not be explained and mathematicians who explained things that could not be observed. James Lighthill In every mathematical investigation, the question will arise whether we can apply our mathematical results to the real world. V.I. Arnold
6.1 Introduction This chapter deals with the theory and applications of kinematic waves to several real-world problems, which include traffic flow on highways, flood waves in rivers, glacier flow, roll waves in an inclined channel, chromatographic models, and sediment transport in rivers. The general ideas and essential features of these problems are of wide applicability. Other applications of conservation laws include various chromatographic models in chemistry and the movement of pollutants in waterways. The propagation of traffic jams is almost similar to the shock waves that cause noise pollution near airports and spaceports. Kinematic wave phenomena also play an important role in traveling detonation and combustion fronts, the wetting water fronts observed in soils after rainfall, and the clanking of shunting trains. All of these problems are essentially based on the theory of kinematic waves developed by Lighthill and Whitham (1955). Many basic ideas and important features of hyperbolic waves and kinematic shock waves are found to originate from gas dynamics, so specific nonlinear models which describe Riemann’s simple waves with Riemann’s invariants and shock waves in gas dynamics are discussed. Considerable attention is also given to nonlinear hyperbolic systems and Riemann’s invariants, generalized simple waves, and generalized Riemann’s invariants. L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_6,
284
6 Kinematic Waves and Real-World Nonlinear Problems
6.2 Kinematic Waves Classical wave motions are described by Newton’s second law of motion together with some reasonable assumptions relating stress to displacement (as in gravity waves), to strain (as in nondispersive longitudinal and transverse waves), or to curvature (as in capillary waves and flexural waves). In contrast with the case of dynamic waves, a class of waves is called kinematic waves when an appropriate functional relation exists between the density and the flux of some physically observed quantity. Kinematic waves are not at all waves in the classical sense, and they are physically quite different from the classical wave motions involved in dynamical systems. They describe, approximately, many important real-world problems including traffic flows on highways, flood waves in rivers, roll waves in an inclined channel, and chromatographic models in chemistry. Lighthill and Whitham (1955) first gave a general and systematic treatment of kinematic waves and applications. In many problems of one-dimensional wave propagation where there is a continuous distribution of either material or some state of the medium, we can define a density ρ(x, t) per unit length and a flux q(x, t) per unit time. Then, we can introduce a flow velocity u(x, t) by u = q/ρ. Assuming that the material (or state) is conserved, we can stipulate that the time rate of change of the total amount in any arbitrary interval x1 ≤ x ≤ x2 must be balanced by the net influx across x1 and x2 . Physically, this states that the quantity in a small length segment changes at a rate equal to the difference between inflow and outflow. Mathematically, this can be formulated as x2 ∂q d x2 dx, (6.2.1) ρ(x, t) dx = q(x1 , t) − q(x2 , t) = − dt x1 ∂x x1 or equivalently,
x2
x1
∂ρ ∂q + ∂t ∂x
dx = 0.
(6.2.2)
If this result is to hold for any arbitrary interval x1 ≤ x ≤ x2 , the integrand must vanish identically, so that ∂q ∂ρ + = 0, (6.2.3) ∂t ∂x provided ρ and q are sufficiently smooth functions. As stated in Chapter 5, (6.2.3) is called the conservation law, kinematic wave equation, or the equation of continuity. Based on theoretical or empirical grounds, we assume that there exists a relation between q and ρ, so that we can write q = q(ρ).
(6.2.4)
Thus, equations (6.2.3) and (6.2.4) form a closed system since there are two equations with two unknown functions. Substituting (6.2.4) in (6.2.3) gives ∂ρ ∂ρ + c(ρ) = 0, ∂t ∂x
(6.2.5)
6.2 Kinematic Waves
where c(ρ) = q (ρ) =
∂q . ∂ρ
285
(6.2.6)
Similarly, multiplying (6.2.3) by c(ρ) leads to the equation ∂q ∂q + c(ρ) = 0. ∂t ∂x
(6.2.7)
This means that q or ρ is constant in waves propagating with velocity c(ρ) given by (6.2.6), and hence, c(ρ) is called the wave propagation velocity. Mathematically, equation (6.2.7) has only one system of characteristics given by dx = c(ρ), dt
(6.2.8)
and along each of these characteristics the flow q or the density ρ is constant. The wave velocity c as given by (6.2.6) is the slope of the flow-density curve for fixed x. In terms of the mean flow velocity u = q/ρ, the wave propagation velocity is given by du d (uρ) = u + ρ . (6.2.9) c = q (ρ) = dρ dρ Thus, c < u when du dρ > 0, that is, the flow velocity increases with density as in < 0, that is, it decreases with density as in flood waves in rivers, and c < u when du dρ traffic flows on highways. Further, the following observations are in order. First, there is one important difference between kinematic waves and dynamic waves. A kinematic wave has only one wave velocity at each point, while dynamic waves possess at least two velocities (forward and backward relative to the medium). Second, kinematic waves are nondispersive, but they suffer from a change in form due to nonlinearity (dependence of the wave speed c on the flow q carried by the wave) exactly as do traveling sound waves of finite amplitude. Consequently, continuous wave forms may develop discontinuities because the faster waves overtake the slower ones. These discontinuities can be described as shock waves because their process of development is identical to that of shock waves in gas dynamics. The law of motion of kinematic shock waves can be derived from the conservation laws, as was the law governing continuous kinematic waves. If the density and flow assume the values ρ1 and q1 on the one side, and ρ2 and q2 on the other side of the shock wave which propagates with velocity U , then the quantity crossing it per unit time can be written either as q1 − U ρ1 or as q2 − U ρ2 . This gives the velocity of the shock waves as (q2 − q1 ) . (6.2.10) U= (ρ2 − ρ1 ) This represents the slope of the chord joining the two points (ρ1 , q1 ) and (ρ2 , q2 ) on the density-flow curve which corresponds to the states behind and ahead of the shock wave when it reaches a given point x. In the limit, when the shock wave become
286
6 Kinematic Waves and Real-World Nonlinear Problems
a continuous wave, the slope of the chord becomes the slope of the tangent, and hence (6.2.10) reduces to (6.2.6). The development of kinematic shock waves was described by examples with figures in Chapter 5. To avoid duplication of a similar analysis, we simply state appropriate results whenever needed. Another important effect, known as diffusion, is confined entirely to the interior of kinematic shock waves, where its effect is crucial in smoothing the nonlinear steeping of the kinematic waves. The effect of diffusion is certainly small outside the shock regions, but inside the shock regions, the effect of diffusion is very important as a second-order approximation. Mathematically, diffusion corresponds to the inclusion in (6.2.7) of an additional term proportional to a second derivative of q. This will happen if the flow-density relation involves some dependence on a derivative of q or ρ in addition to q, ρ, and x. We assume that, for each x, q is a function of ρ and ρt ; alternatively, ρ is a function of q and qx because ρt = −qx . Substituting for ρ in (6.2.3), we obtain ∂ρ ∂q ∂ρ ∂ 2 q ∂q + + = 0. (6.2.11) ∂q ∂t ∂qx ∂x∂t ∂x Further, if the coefficients of the derivatives of q in (6.2.11) are approximated as functions of q and x alone, then equation (6.2.11) becomes ∂q ∂2q ∂q + c(ρ) +ν = 0, ∂t ∂x ∂x∂t
(6.2.12)
∂q ∂ρ where c = ∂ρ is the kinematic wave velocity as before and ν = c( ∂q ). This is a x typical nonlinear equation representing diffusive kinematic waves. Invoking the first approximation qt ∼ −cqx , (6.2.12) may be rewritten as
∂q ∂2q ∂q + c(ρ) = νc 2 . ∂t ∂x ∂x
(6.2.13)
Physically, there is no difference between the two equations (6.2.12) and (6.2.13). However, further details of the diffusion equation will be pursued in Chapter 8. The rest of this chapter is devoted to a detailed treatment of several real-world problems as applications of kinematic waves, Riemann’s simple waves, and Riemann’s invariants and their extensions.
6.3 Traffic Flow Problems Traffic flow on a highway is one of the most common real-world problems. Based on the works of Lighthill and Whitham (1955) and Richards (1956), we consider the traffic flow on a long highway under the assumptions that cars do not enter or exit the highway at any one of its points and that individual cars are replaced by a continuous density function. We take the x-axis along the highway and assume the traffic flows in the positive direction. This problem can be described by three fundamental traffic variables: the traffic density ρ(x, t), which is equal to the number of cars per unit length at position x and at time t, the traffic flow q(x, t), the number of cars passing a
6.3 Traffic Flow Problems
287
Fig. 6.1 (a) Velocity–density curve and (b) flow–density curve.
fixed point in unit time, and the traffic velocity u(x, t), which represents the velocity of a car. We use the theory of kinematic waves to formulate the problem in terms of a first-order, nonlinear, partial differential equation on the basis of conservation of cars and experimental relationships between the car velocity and traffic density. So, the equations governing the traffic flow are ∂q ∂ρ + = 0, ∂t ∂x q = q(ρ),
(6.3.1) (6.3.2)
where q is some function of ρ as determined by theoretical or experimental findings of traffic flow phenomena. The functional relation (6.3.2) seems to be reasonable in the sense that the density of cars surrounding a given car indeed controls the speed of that car. In fact, this relation depends on other factors including speed limits, weather conditions, and road conditions. Several specific relations have been suggested by Haight (1963). On the basis of observations of traffic flow, we make a basic simplifying assumption that the velocity of a car at any point along the highway depends only on the traffic density, that is, u = U (ρ), and hence, the traffic flow q = ρU (ρ). Clearly, u(ρ) must be a monotonically decreasing function of density ρ. If there are no cars on the highway (corresponding to very low densities), then the car would travel at a finite maximum speed umax , that is, U (0) = umax = um . As the density of cars increases, the velocities of the cars would continue to decrease, and hence, U (ρ) = 0 as ρ → ρmax = ρm , where ρm is the maximum traffic density corresponding to what is called bumper-to-bumper traffic. Also, it follows that du dρ = U (ρ) ≤ 0. The decreasing feature of traffic velocity is shown in Figure 6.1(a). On the other hand, the important feature of traffic flow Q(ρ) = ρU (ρ) can be inferred from U (ρ). Clearly, Q(ρ) → 0, as ρ → 0, and ρ → ρmax = ρm (U (ρm ) = 0). This means that the traffic flow Q(ρ) is an increasing function of density ρ until it attains a maximum value Qmax = qM for some ρ = ρM in 0 < ρ < ρm . In general, Q(ρ) assumes a parabolic form, concave downward, Q (ρ) < 0. All of these features of the flow–density curve are shown in Figure 6.1(b). It is important to point out that, corresponding to any point on the flow–density curve, the mean speed U = Q/ρ = tan θ of cars represents the slope of the chord from the origin as shown in Figure 6.1(b).
288
6 Kinematic Waves and Real-World Nonlinear Problems
dq The propagation speed c = dρ of density waves carrying continuous changes of flow through the streams of cars is the slope of the tangent to the curve at that point. This slope is smaller provided that the mean speed decreases with increased density. Hence, we can write d du dq = (ρu) = u + ρ . (6.3.3) c= dρ dρ dρ
Evidently, c < u if du dρ < 0, and c = u only at a very low density. Usually, under light traffic conditions, there are few cars on the highway traveling at high speed. On the other hand, under heavy traffic conditions, there are many cars on the road, and these cars travel slowly. So again, traffic flow would be small. Substituting q = U (ρ) into (6.3.1) gives the first-order, nonlinear, partial differential equation for traffic flow in the form ∂ρ ∂ρ + c(ρ) = 0, ∂t ∂x where
(6.3.4)
c(ρ) = Q (ρ) = U (ρ) + ρ U (ρ)
(6.3.5)
represents the velocity of the density waves. Since U (ρ) < 0, c(ρ) < U (ρ), that is, the velocity of these waves is always less than that of cars, and drivers experience such waves and are warned of disturbances ahead. Figure 6.1(b) shows that c(ρ) = Q (ρ) > 0 for all ρ in [0, ρM ), is zero at ρ = ρM , and then, is negative in (ρM , ρm ]. All these cases mean that waves propagate forward relative to the highway in [0, ρM ], are stationary at ρ = ρM , and then, travel backward in [ρM , ρm ]. Further, discontinuous waves are likely to occur on any segment of highway when the traffic is very heavy in front and light behind. For waves on which the traffic flow is less dense, cars travel forward faster than, and hence tend to catch up with, those on which the flow is more dense. When this happens, a group of continuous waves can coalesce into discontinuous (or shock) waves. According to the theory of characteristics of the first-order nonlinear equation, the characteristic equations for (6.3.4) are dx dρ dt = = , 1 c(ρ) 0 with the solution
(6.3.6)
dx = c(ρ). (6.3.7) dt Since c(ρ) is constant when ρ is constant, the characteristics are straight lines in the (x, t)-plane, and the characteristic velocity c(ρ) = Q (ρ) is the slope of the tangent to the curve at a point. Actual observational data of traffic flow indicate that typical results on a single lane highway are ρm ≈ 225 cars per mile, ρM ≈ 80 cars per mile, and qM ≈ 1590 cars per hour. Thus, the maximum traffic flow qM occurs at a low speed u = qM /ρM ≈ 20 miles per hour. ρ = const.
on
6.3 Traffic Flow Problems
289
Fig. 6.2 Multi-valued density profile at t = t2 t1 > t0 > 0.
Next, we solve the nonlinear traffic flow equation (6.3.4) with the initial and boundary conditions ρ(x, 0) = f (x), x ∈ R, (6.3.8) where f (x) vanishes outside a finite domain of x and ρ(x, t) → 0, as |x| → ∞. This initial-value problem has already been solved in Section 5.2, and hence, we state the result without proof. The solution is given by ρ(x, t) = f (ξ), where
x = ξ + t F (ξ),
F (ξ) = c f (ξ) = c(ρ).
(6.3.9ab) (6.3.10)
The velocity c(ρ) = Q (ρ) is the slope of the traffic flow curve drawn in Figure 6.1(b), so that the density waves move forward or backward relative to the highway depending on whether ρ < ρM or ρ > ρM . We consider an initial density curve ρ(x, 0) = f (x) in the form of a hump which has a maximum ρm at some point x, as shown in Figure 6.2 so that ρm > ρ0 and c(ρ0 ) < c(ρm ). Thus, the point on the hump with density ρm travels slower than the point with density ρ0 . As time progresses, the density profile continues to steepen at the back and flatten at the front, eventually leading to a multi-valued solution which is physically inadmissible. In other words, a vertical segment develops at the back for t2 t1 . The vertical segment corresponds to a point in the (x, t)-plane at which the density is discontinuous and, therefore, at which at least one of the partial derivatives ρx and ρt no longer exists. Thus the distortion of the initial density profile and the development of discontinuity as a shock are remarkable features of nonlinear traffic flow phenomena. However, the solution incorporating the discontinuity must satisfy conservation laws, and hence, the total number of cars in any interval (x1 , x2 ) represented by the integral of ρ with respect to x from x1 to x2 (> x1 ) must be unchanged. Thus, the shock must be inserted at that value of x which leaves the total area under the curve unchanged, that is, at the point S such that the area ABC = area CDE. This simple result that the two areas cut off by the vertical line through
290
6 Kinematic Waves and Real-World Nonlinear Problems
Fig. 6.3 (a) Initially heavier traffic ahead (ρ2 > ρ1 ) and (b) intersecting characteristics.
Fig. 6.4 Graphical representation of solution.
S are equal is known as Whitham’s rule. So, the Whitham rule of equal area can be used to determine the position of the shock path for certain nonlinear equations. We consider a situation in which traffic initially becomes heavy further along the highway. As already indicated in Section 5.2, the solution represents a compression wave. It is convenient to assume the initial data, as shown in Figure 6.3(a), ρ1 if x < 0, (6.3.11) ρ(x, 0) = ρ2 if x > 0, where ρ1 and ρ2 are constants, 0 ≤ ρ1 < ρ2 ≤ ρmax , and hence, U1 > U2 . The density wave for the lighter travel with velocity c(ρ1 ) = Q (ρ1 ), which is greater than the velocity c(ρ2 ) = Q (ρ2 ) of the heavier traffic density wave, that is, c(ρ1 ) > c(ρ2 ). A set of characteristics using the initial data (6.3.11) is shown in Figure 6.3(b). In any situation where the traffic becomes denser further along the road, characteristics intersect. At any point where two characteristics intersect, ρ = ρ1 and ρ = ρ2 , but it is physically impossible for the traffic density to be multi-valued. A jump occurs across a curve as a shock. The behavior of the ‘jump’ in the density ρ in the (x, t)-plane just treated indicates that a shock should be inserted between two regions of constant density. Since the overlapping starts at the origin, we look for a solution of the form shown in Figure 6.4, two regions of constant density ρ separated by a constant shock velocity given by (5.3.6). It can be shown that equation (6.3.4) and the initial data (6.3.11) are satisfied on both sides of the shock and the condition (5.3.6) is satisfied on the shock. Evidently, ρ = ρ1 = const. satisfies equation (6.3.4)
6.3 Traffic Flow Problems
291
Fig. 6.5 The solution with a fan centered at the origin.
and (6.3.11) for x < 0 at t = 0; similarly, ρ = ρ2 = const. for x > 0 at t = 0. The shock condition (5.3.6) is satisfied if Q1 − Q2 , (6.3.12) U= ρ1 − ρ2 that is, if the shock propagates at the constant velocity given by (6.3.12). The shock path is a straight line through the origin in the (x, t)-plane. Thus the traffic density ρ(x, t) satisfies all equations and initial data and is single-valued for all x and t except across the shock. There is another case associated with the initial data which corresponds to ρ1 > ρ2 , and hence, U1 < U2 , and c(ρ1 ) < c(ρ2 ). Physically, this case represents a situation in which the traffic initially becomes light further along the highway. Using the initial conditions (6.3.11), characteristics are drawn in Figure 6.5. It is important to point out that there are no characteristics within the infinite sector AOB, and hence, no solution for the traffic density ρ(x, t). According to Section 5.4, there are two regions, each of constant density ρ, separated by a fan centered at the origin. The solution is given by ⎧ to the left of OA, ⎪ ⎨ ρ1 to the right of OB, ρ(x, t) = ρ2 ⎪ ⎩ ρ0 , const., ρ2 < ρ0 < ρ1 on x = q (ρ0 ) in the sector AOB. (6.3.13) The solution in the sector AOB is the characteristic form, equation (6.3.7), of the solution of equation (6.3.4). Across OA and OB, ρ(x, t) is continuous, but ρx and ρt are discontinuous (ρx and ρt are nonzero in the sector AOB but zero elsewhere). Thus, the problem is solved for both cases: (i) ρ1 < ρ2 and (ii) ρ1 > ρ2 . However, a remarkable fact involved in this problem with case (ii) is that there exists another discontinuous solution which satisfies all equations and initial conditions. This solution consists of a shock dividing two regions of constant density ρ. All characteristics on the left of the shock are parallel, and so are those to the right of the shock as shown in Figure 6.6.
292
6 Kinematic Waves and Real-World Nonlinear Problems
Fig. 6.6 Characteristics and the discontinuous solution.
For case (ii), there are at least two solutions: (a) the fan-centered solution and (b) the shock solution. Both satisfy all equations and the initial conditions, but the solution is not unique. The main question is to find which one of these solutions represents a physically meaningful solution. This question was answered in Section 5.4 by introducing the Lax entropy criterion (5.4.17). Example 6.3.1 The Green Signal Problem in Front of the Traffic. We consider a stream of cars stopped by a red signal at x = 0. The road is jammed initially behind the signal and there is no traffic ahead of the signal. As soon as the signal turns green, the stream of cars starts moving across x = 0. The initial state of the traffic is given by ρm if x < 0, ρ(x, 0) = ρm H(−x) = (6.3.14) 0 if x > 0, where ρm is the maximum density and H(x) is the Heaviside unit step function. We assume that the traffic flow q(ρ) is quadratic in the region 0 < ρ < ρm and zero otherwise, that is, qm ρ q= 1− ρ, 0 < ρ < ρm . (6.3.15) ρm ρm It is noted that the maximum flow rate is 14 qm which occurs at ρ = 12 ρm . The velocity of the traffic flow is u = ρq , that is, ρ , u = um 1 − ρm
(6.3.16)
where um = (qm /ρm ) is the maximum speed when ρ = 0. The traffic flow equation for the density ρ(x, t) is given by ρt + c(ρ)ρx = 0, where
(6.3.17)
6.3 Traffic Flow Problems
293
Fig. 6.7 Characteristics due to a discontinuous initial density.
2ρ . c(ρ) = q (ρ) = um 1 − ρm
(6.3.18)
Thus, the characteristic equations for (6.3.17) are dt dx dρ = = , 1 c(ρ) 0 where dρ = 0,
2ρ dx . = c(ρ) = um 1 − dt ρm
Thus the density ρ remains constant along a characteristic given by dx 2ρ . = um 1 − dt ρm
(6.3.19ab)
(6.3.20)
Any characteristic that intersects the positive x-axis at x = x0 > 0 has the slope
2ρ 2ρ(x, 0) dx = um 1 − = um . = um 1 − (6.3.21) dt ρm ρm This gives a family of characteristics as the right-leaning straight lines, as shown in Figure 6.7, x = um t + x0 . (6.3.22) On the other hand, any characteristic that intersects the negative x-axis at x = x0 < 0 has the slope
dx 2um = −um . = um 1 − (6.3.23) dt um As shown in Figure 6.7, equation (6.3.23) gives a family of characteristics representing the left-leaning straight lines x = −um t + x0 .
(6.3.24)
294
6 Kinematic Waves and Real-World Nonlinear Problems
Fig. 6.8 Traffic density after the signal turns green.
In the fan-like region −um t < x < um t, all characteristics must pass through the x origin and must be straight lines. Hence, dx dt = t . It follows from (6.3.20) that 2ρ x . (6.3.25) = um 1 − t ρm Solving for ρ gives ρ=
x 1 , ρm 1 − 2 um t
−um t < x < um t.
(6.3.26)
At any instant, the density varies linearly in x from ρm at x = −um t to zero at x = um t. Thus, the solution (6.3.26) for ρ(x, t) for all x at any time t is drawn in Figure 6.8. Example 6.3.2 The Traffic Signal Problem. We consider the same problem as in Ex3 (200 − ρ)ρ, and hence, the velocity of the ample 6.3.1 with the traffic flow q = 10 3 traffic density waves is c(ρ) = 5 (100 − ρ). We have to find the traffic density when the traffic signal turns green at t = 0, so that the initial density is given by (6.3.14). The typical value of ρm varies from 200 to 225. According to the theory described above, the solution of the initial-value problem is f (x, t) = f (ξ) = ρm H(−ξ),
3 x = ξ + c f (ξ) = ξ + 100 − f (ξ) t. 5 (6.3.27ab)
Using ρm = 200, we obtain the solution as 200 if ξ < 0, −|ξ| − 60t if ξ < 0, f (x, t) = x= 0 if ξ > 0, ξ + 60t if ξ > 0.
(6.3.28ab)
These two results can be combined to obtain the traffic density for any t > 0
6.3 Traffic Flow Problems
ρ(x, t) =
200
if x < −60t,
0
if x > 60t.
295
(6.3.29)
However, the solution for ρ(x, t) is not known in the interval −60t < x < 60t. The nonexistence of the solution in this interval is somewhat unusual. It is possibly due to the fact that the initial condition of the traffic flow is discontinuous at ξ = 0, but the use of the conservation law (6.3.4) requires that ρ(x, t) be differentiable. Example 6.3.3. Obtain the traffic density ρ(x, t) governed by (6.3.4) with the given 3 (200 − ρ)ρ and with the initial state of the traffic flow as traffic flow q(ρ) = 10 ρ(x, 0) = f (x) = 1 − x.
(6.3.30)
In this problem, the velocity of the density waves is given by c(ρ) =
3 dq = (100 − ρ). dρ 5
(6.3.31)
The solution of the initial-value problem is given by (6.3.9ab), that is, ρ(x, t) = f (ξ), x = ξ + t c f (ξ) ,
(6.3.32a) (6.3.32b)
where c{f (ξ)} = 35 (100 − 1 + ξ) = 35 (99 + ξ). Solving for ξ from (6.3.32b) gives ξ=
(x − 35 · 99t) . (1 + 35 t)
(6.3.33)
This result can be used to eliminate ξ from (6.3.32a) so that the solution becomes ρ(x, t) = (1 − ξ) =
(1 − x + 60t) . (1 + 3t 5)
(6.3.34)
This traffic flow model is based on the first-order approximation, and hence, the original assumptions that q = Q(ρ) and u = U (ρ) are not good approximations. For a better approximation, it is reasonable to assume that q and u are functions of both ρ and its gradient ρx . Hence, we can assume that q = Q(ρ) − νρx , ρx , u = U (ρ) − ν ρ
(6.3.35a) (6.3.35b)
where ν is a positive constant in the context of traffic flow because drivers are supposed to reduce their speed to account for heavy traffic ahead. This introduces two additional effects: (i) diffusion of the waves and (ii) response time defined by the time lag in the response of the driver and his or her car to any changes in traffic conditions.
296
6 Kinematic Waves and Real-World Nonlinear Problems
To incorporate the first effect of diffusion produced by the driver’s response to conditions ahead, we substitute (6.3.35a) in the traffic flow equation (6.3.1), which gives a new parabolic equation in the form ρt + c(ρ)ρx = νρxx .
(6.3.36)
This equation is known as the Burgers equation, which will be discussed in great detail in Chapter 8. Physically, the second-order term introduced by the diffusion will smooth out the traffic flow. In other words, the effect of diffusion tends to eliminate any discontinuity involved in the solution of the original hyperbolic traffic flow equation (6.3.4). To incorporate the second effect, the following equation: ρx 1 Du = ut + uux = − u − U (ρ) + ν (6.3.37) Dt τ ρ can be introduced for the acceleration of the car where the coefficient τ is a measure of the response time. Equation (6.3.37) is to be solved combined with the conservation law ρt + (ρu)x = 0. (6.3.38) For the case of small ν and τ , (6.3.37) can be approximated by the original first-order result, that is, by u = U (ρ). Before we solve the nonlinear problem, it would be helpful to examine the linearized version of equations (6.3.37) and (6.3.38). These equations can be linearized for small perturbations about the equilibrium state ρ = ρ0 and u = u0 = U (ρ0 ), so that we can introduce ρ = ρ0 + ρ˜(x, t) and u = u0 + u ˜(x, t). We substitute these results in (6.3.37) and (6.3.38) and retain only first-order terms to obtain ν ρ˜ , (6.3.39) τ [˜ ut + u0 u ˜x ] = − u ˜ − U (ρ0 )˜ ρ+ ρ0 ρ˜t + u0 ρ˜x + ρ0 u ˜x = 0. (6.3.40) The propagation velocity of the density wave is represented by c0 = ρ0 U (ρ0 ) + U (ρ0 ),
(6.3.41)
and hence, U (ρ0 ) = −(u0 − c0 )/ρ0 which is used in (6.3.39), (6.3.40) to eliminate u ˜, and the final form of the equation for ρ˜(x, t) is ∂ ∂ ∂ ∂ (˜ ρt + c0 ρ˜x ) = ν ρ˜xx − τ + u0 + u0 ρ˜. (6.3.42) ∂t ∂x ∂t ∂x It is noted that the first term on the right-hand side represents diffusion, whereas the second term arises from the effect of the response time τ . If ν and τ are very small, (6.3.42) reduces to a simple first-order equation. ρ˜t + c0 ρ˜x = 0.
(6.3.43)
6.4 Flood Waves in Long Rivers
297
The solution of this equation is ρ˜ = f (x − c0 t) representing waves traveling with speed c0 . Equation (6.3.43) suggests that ∂ ∂ ≈ −c0 . ∂t ∂x
(6.3.44)
This approximation is utilized in the right-hand side of (6.3.42), so that it reduces to the form ∂ 2 ρ˜ (˜ ρt + c0 ρ˜x ) = ν − τ (u0 − c0 )2 . (6.3.45) ∂x2 It can be inferred from this equation that the solution would be stable or unstable depending on whether ν > τ (u0 − c0 )2
or ν < τ (u0 − c0 )2 .
(6.3.46ab)
We examine the stability of the plane wave solutions of the original equation (6.3.42) without any approximation. This equation admits solutions of the form ρ˜(x, t) = a exp i(kx − ωt) , (6.3.47) provided the equation ν(ω − u0 k)2 + i(ω − c0 k) − νk 2 = 0
(6.3.48)
is satisfied. Thus, the solutions would be stable if I(ω) < 0 for both roots of (6.3.48). Finally, we observe that the right-hand side of (6.3.42) is a wave operator so that it corresponds to factoring as ∂ ρ˜ ∂ ∂ ∂ ∂ ∂ ρ˜ ρ˜, (6.3.49) + c0 = −τ + c+ + c− ∂t ∂x ∂t ∂x ∂t ∂x
where c+ = u0 +
ν , τ
c− = u 0 −
ν . τ
(6.3.50ab)
Evidently, one wave propagates with the speed c+ and the other with the speed c− < c+ . Thus, for small τ , the equation (6.3.49) can be approximated by ρ˜t + c0 ρ˜x = 0,
(6.3.51)
provided c− < c0 < c+ , which is exactly the stability criterion (6.3.46ab). Thus, the conclusion is that the solution is stable provided that the condition c− < c0 < c+ holds.
6.4 Flood Waves in Long Rivers Another application of the theory of kinematic waves dealing with flood waves in long rivers will be presented here in some detail. We consider the propagation of
298
6 Kinematic Waves and Real-World Nonlinear Problems
flood waves in a long rectangular river of constant breadth. We take the x-axis along the river which flows in the positive x-direction and assume that the disturbance is approximately the same across the breadth. In this problem, the depth h(x, t) of the river plays the role of density involved in the traffic flow model discussed in Section 6.3. We denote the volume flow per unit breadth and per unit time by q(x, t). An argument similar to that of Section 6.2 gives the kinematic wave equation for the flow in a river as ∂q ∂h + = 0. (6.4.1) ∂t ∂x Although the flow in a river is fairly complicated, we assume a simple functional relation q = Q(h) as a first approximation to increased flow as the water level rises. Thus, equation (6.4.1) becomes ∂h ∂h + c(h) = 0, ∂t ∂x
(6.4.2)
where c(h) = Q (h) and Q(h) is essentially determined from the balance between the gravitational force and the frictional force of the river bed. One such result for Q(h) is the famous Chézy formula given by Q(h) = h u,
(6.4.3)
√ where √ u is the average velocity of the river flow and it is proportional to h, so that u = α h, where α is a constant. The propagation velocity of flood waves, by (6.2.6), is given by c(h) = Q (h) =
du 3 √ d 3 (uh) = u + h = α h = u. dh dh 2 2
(6.4.4)
This clearly shows that the flood waves propagate one and a half times faster than the stream velocity. A more general result is u = αhn for a constant α so that the volume flow is given by (6.4.5) Q(h) = hu = αhn+1 . Thus the propagation velocity of flood waves is c(h) = Q (h) = (n + 1)u.
(6.4.6)
This also confirms the fact that flood waves propagate faster than the fluid velocity for all positive n. When n = 23 , (6.4.5) is called the Manning formula. When the kinematic wave theory described in Section 5.3 is applied to flood waves in rivers, kinematic shock waves play a major role in the forward regions of flood waves. The general solution may be taken from Section 5.3 with shocks fitted in as discontinuities which satisfy the condition U=
(u2 h2 − u1 h1 ) . (h2 − h1 )
(6.4.7)
6.4 Flood Waves in Long Rivers
299
Fig. 6.9 Monoclinal flood wave.
The process by which kinematic waves may steepen into shock waves with a considerable change in flow in a relatively short distance has not been very clearly described in the literature. However, the possibility that such a wave propagates down a river, with different, uniform flows upstream and downstream of it, has been visualized as a flood wave model. Such a wave is called the monoclinal flood wave (or the steady wave profile). In fact, this is a progressing wave whose profile tends to different constant states upstream and downstream, with lower depth downstream, joined by a steadily falling region, as indicated schematically in Figure 6.9. It has been suggested by Whitham (1974) that discontinuities described by the jump conditions in hydraulic theory are in reality the turbulent bores familiar in water wave phenomena as hydraulic jumps or breakers on a beach. Although the above nonlinear model based on a first-order approximation, gives a fairly good description of the flood wave phenomena, Seddon (1900) raised a number of questions about the formulation of any river flow model based solely on the Chézy relation or its extensions. First, rivers do not have a uniformly sloping bed, nor do they, in any way, approximate this situation. Second, the river bed constantly varies with time, since its material is readily handled by the flow. To improve the above model, it is reasonable to assume that the volume flow depends on both h and x so that q = q(h, x). This relation can be combined with (6.4.1) to give a first approximation for unsteady flows which vary slowly. Then, h(x, t) satisfies the equation ∂q ∂h ∂q ∂h + · = . (6.4.8) ∂x ∂h ∂x ∂x ∂q )x constant = c(h, x), we obtain Multiplying (6.4.1) by c = ( ∂h ∂q ∂q + c(h, x) = 0. ∂t ∂h
(6.4.9)
This means that q is constant for waves traveling past the point with velocity c(h, x). Mathematically, the equation has only one system of characteristics given by dx dt = c, and the flow q is constant along each of these characteristics. The wave velocity c = c(h, x) is the slope of the density-flow curve for fixed x. This result for flood waves is known as the Kleitz–Seddon law.
300
6 Kinematic Waves and Real-World Nonlinear Problems
6.5 Chromatographic Models and Sediment Transport in Rivers In chemistry, chromatography is described as an exchange process between a solid and a fluid. In general, the exchange process may involve particles or ions of some substance or it may be heat exchange between the solid bed and a fluid. The general problem is extremely complicated. However, it is possible to develop a simple one-dimensional mathematical model for single-solute chromatography based on the assumption of a local equilibrium. The concentration of solute per unit volume in the fluid phase is denoted by c(x, t), and the concentration of solute in the solid phase by n(x, t). A simple argument of the balance of influx and outflux leads to the basic conservation equation ε
∂c ∂n ∂c + εu + (1 − ε) = 0, ∂t ∂x ∂t
(6.5.1)
where ε is a constant fractional void volume in the bed and u is a constant interstitial velocity of fluid through the bed. The equilibrium equation is (6.5.2)
n = f (c),
where f (c) is assumed to be positive for physical considerations, but f (c) in general depends on temperature. Often, f (c) is called an absorption isotherm. Substituting (6.5.2) in (6.5.1) gives
Introducing ν =
ε + (1 − ε)f (c)
(1−ε) ε
∂c ∂t
+ εu
∂c = 0. ∂x
(6.5.3)
in (6.5.3) yields ∂c ∂c 1 + ν f (c) +u = 0. ∂t ∂x
(6.5.4)
Finally, in terms of a new function defined by g(c) = c + ν f (c), equation (6.5.4) assumes the form ∂c ∂c = 0. (6.5.5) g (c) + u ∂t ∂x The function g(c) is called the column isotherm. This equation is somewhat similar to that of the traffic flow equation. So, we may skip the detailed method of solution and state the salient features of the solution. It is convenient to introduce nondimensional variables defined by x∗ =
x ,
t∗ =
ut ,
(6.5.6ab)
where is a characteristic length for the column. Dropping the asterisks, equation (6.5.5) becomes ∂c ∂c g (c) + = 0. (6.5.7) ∂t ∂x
6.5 Chromatographic Models and Sediment Transport in Rivers
301
This nonlinear chromatographic equation can be solved subject to the initial and boundary data c(x, 0) = φ(x), c(0, t) = ψ(t), (6.5.8ab) where φ and ψ are arbitrary functions. It is often convenient to express the initial and boundary data in the parametric form x = ξ, t = τ,
c = φ(ξ) at t = 0, c = ψ(τ ) at x = 0.
(6.5.9) (6.5.10)
The characteristic equation of (6.5.7) is dt = g (c), dx
(6.5.11)
so that g (c) = νf (c). For convex isotherms, f (c) < 0, and hence, g (c) < 0. Next, we solve a particular model, called the Langmuir isotherm model, described by αc N κc = = f (c), (6.5.12) n= (1 + κc) (1 + κc) where N is the saturation concentration, κ is a rate constant, and α = N κ. In this case, the characteristic equation is να dt . = g (c) = 1 + ν f (c) = 1 + dx (1 + κc)2
(6.5.13)
We consider a particular case of (6.5.9), (6.5.10) representing a linear distribution of solute over a finite segment of the x-axis so that c0 ξ if 0 ≤ ξ ≤ a, α x = ξ, c = at t = 0, (6.5.14) c0 if ξ ≥ a t = τ, c = 0 at x = 0, (6.5.15) where c0 is constant. The graphs of the Langmuir isotherm and the characteristics in the (x, t)-plane are drawn in Figure 6.10(a) and (b). It follows from Figure 6.10(b) that all characteristics originating from the t-axis have slope g (0) = 1 + να, whereas those from the x-axis for ξ ≥ a are less steep with a slope g (c0 ). These two sets of characteristics yield two constant states, c = 0 and c = c0 , respectively. On the other hand, the characteristics originating from the interval 0 ≤ ξ ≤ a on the x-axis have slope between g (0) and g (c0 ), and hence, they represent a fan or expansive wave. Thus, the solution consists of two constant states separated by the expansion wave. For any point in the expansion wave region, we obtain
a να , ξ(c) = c. (6.5.16ab) t = x − ξ(c) 1 + 2 (1 + κc) c0
302
6 Kinematic Waves and Real-World Nonlinear Problems
Fig. 6.10 (a) The Langmuir isotherm and (b) characteristics in (x, t)-plane.
In the limit as a → 0, the characteristics originating from the interval 0 ≤ ξ ≤ a tend to translate to the left, but their slopes remain unchanged. In fact, these characteristics all originate from the origin, that is, the simple waves become centered there. In reality, this represents the elution of a uniformly absorbed column, which is a Riemann problem. In this case (a → 0), we obtain να t = g (c) = 1 + x (1 + κc)2
(6.5.17)
from any point in the simple wave region, so that (κc) can be expressed as a function of ( xt ) only: − 12 √ t −1 . (6.5.18) (κc) = −1 + να x This result was recognized in the theory of chromatography as early as 1945. We next consider the saturation of a clean column with the conditions c(x, t) = 0 at t = 0 for x > 0, (6.5.19ab) c(x, t) = c0 at x = 0 for t > 0. Obviously, the characteristics originating from the x-axis (t = 0) have slopes g (0) = (1+να), and they are steeper than those originating from the t-axis (x = 0), since the slopes of the latter have slopes g (c0 ) = 1 + (να)/(1 + κc0 )2 . The characteristics are shown in Figure 6.11. It is important to point out that there is a characteristic corresponding to every concentration between 0 and c0 generating from the origin. Three characteristics PA, PO, and PB corresponding to concentrations 0, 1 1 [−1 + {να/( xt − 1)} 2 ], and c0 , respectively, intersect at P . Hence, there are three κ values of c at P showing the multi-valuedness of the concentration function c. This physically unacceptable situation arises from the fact that the higher concentrations move faster than the lower ones. The concentration wave propagating with a sloping front leads to a compression wave which gets progressively steeper as the higher concentration overtakes the lower, until the wave overtops itself as shown in Figure 6.12. Ultimately, the wave form breaks to give a multi-valued solution for c(x, t). In other words, the solution develops a discontinuity. In contrast to linear waves, the breaking phenomenon is typical of nonlinear waves.
6.5 Chromatographic Models and Sediment Transport in Rivers
303
Fig. 6.11 Overlapping characteristic lines.
Fig. 6.12 Distortion of the waveform.
Another example of the application of kinematic waves was given by Kynch (1952). The process of sedimentation of solid particles in a fluid is one of great practical importance. It is based on an assumption that the velocity u of a particle is a function of only the local concentration ρ of particles in its immediate neighborhood. It is also assumed that the particles have the same size and shape. The particle flux S = ρu represents the number of particles crossing a horizontal section per unit area per unit time. Therefore, the particle flux S at any level determines or is determined by the particle concentration. As ρ increases from zero to its maximum value ρm , the velocity u decreases continuously from a finite value u to zero. The variation of S is very complicated, but a simple model is useful for understanding the key features of the problem. We denote the height of any level above the bottom of the column of dispersed particles by x. Based on the argument that the accumulation of particles between two adjacent layers at x and x + dx is the difference between the inflow and the outflow, the continuity equation is given by ∂S ∂ρ = . ∂t ∂x
(6.5.20)
∂ρ ∂ρ + c(ρ) = 0, ∂t ∂x
(6.5.21)
Or equivalently,
where
304
6 Kinematic Waves and Real-World Nonlinear Problems
c(ρ) = −S (ρ) = − u(ρ) + ρ u (ρ) .
(6.5.22)
Equation (6.5.21) is the kinematic wave equation, and its mathematical analysis would be similar to that of Section 6.2. Clearly, the characteristics of (6.5.21) are given by dx = c(ρ). dt
(6.5.23)
As ρ and, hence, c(ρ) are constant, the characteristics are straight lines. The basic equation for the sediment transport of solid particles is similar to the kinematic wave equation. So the results of the kinematic wave theory developed in Section 6.2 can be applied to sediment transport phenomena. Without any further discussion of mathematical analysis, we simply mention some key features of this problem. Any discontinuous changes in the particle concentration can occur in two ways: (i) a discontinuity of the first kind in the particle concentration is due to a sudden finite change of concentration at a certain level, and (ii) a discontinuity of the second kind is due to a very small change in the particle concentration. If ρ1 and u1 represent the concentration and the velocity of particles above the discontinuity and ρ2 and u2 denote the same quantities below the discontinuity, the velocity of the shock waves U is obtained from the equation ρ1 (u1 + U ) = ρ2 (u2 + U ), so that U=
(S1 − S2 ) (u1 ρ1 − u2 ρ2 ) = . (ρ2 − ρ1 ) (ρ2 − ρ1 )
(6.5.24)
(6.5.25)
In a diagram in the (ρ, S)-plane, velocity U represents the slope of the chord joining the points (ρ1 , S1 ) and (ρ2 , S2 ). On the other hand, the discontinuity of the second kind gives ρ2 − ρ1 = dρ where dρ is small so that (6.5.25) becomes U = − dS dρ = c(ρ). This means that the wave velocity c is equal to the shock wave velocity between concentrations ρ and ρ + dρ. Evidently, a small change dρ, if maintained, is propagated through a dispersion of concentration ρ with velocity c just as sound waves propagate through air with a definite velocity. Finally, for dispersions where the concentration increases downward toward the bottom, the condition for the formation of a first-order discontinuity can be expressed in the following equivalent ways: (a) the characteristics in the (x, t)-plane would intersect, (b) the velocity of propagation c(ρ) increases with concentration, or (c) the flow-concentration curve represented by S = S(ρ) in the sediment transport phenomenon is concave to the ρ-axis. If c(ρ) increases with ρ, small concentration changes from denser regions below move faster upward than those in the less dense regions above and, subsequently, overtake them. This means that the concentration gradient increases until a firstorder discontinuity is developed. On the other hand, if c(ρ) decreases with ρ, the concentration gradient decreases, and hence, any discontinuity will be disseminated.
6.6 Glacier Flow
305
6.6 Glacier Flow A glacier is a huge mass of ice that flows over the land under its own weight. Mathematically, a glacier can be treated, essentially, as a one-dimensional flow system, which continuously either gains new material by snowfall or loses it by melting and evaporation. The following discussion on waves in glaciers is based on papers by Nye (1960, 1963), who developed a modern theoretical analysis of glacier flow. We consider the model of a glacier (or a large ice sheet) which rests on an inclined plane of slope α and flows down the plane in the x-direction. We also assume that the glacier is of unlimited extent in the direction normal to x, and α = α(x) is a slowly varying function of x. The thickness of the glacier is assumed as h = h(x, t) such that (hx )t 1. Then, the surface slope β(x, t) downhill in the x-direction is given by (6.6.1) β(x, t) = α − hx . We denote the volume of ice by q(x, t) at a given point x, in unit time, and in unit breadth. This quantity q(x, t) is called a flow. The conservation of ice volume gives the equation ∂q ∂h + = a(x, t), (6.6.2) ∂t ∂x where a(x, t) is the rate of accumulation of ice at the surface, that is, the rate of addition of ice to the upper surface by snowfall and avalanching, measured as thickness of ice per unit time. The negative value of a corresponds to melting or evaporation of ice from the upper or lower surface of the glacier. Assuming that q is a given function of x and h and, then, following Lighthill and Whitham (1955), we multiply ∂q )x to obtain the equation for the glacier flow equation (6.6.2) by c = ( ∂h ∂q ∂q +c = ac. ∂t ∂x
(6.6.3)
If a = 0, q would be constant in the characteristics, that is, there exist kinematic waves of constant flow q traveling with velocity c. In general, c differs from the average velocity of the ice itself, which is given by u = hq . To examine the instability of the compression region of a glacier flow, we assume that q = q(x, h, α). Multiplying ∂q )x,α gives equation (6.6.2) by c = ( ∂h ∂q ∂q ∂h c + = ac, (6.6.4) · ∂x ∂h x,α ∂t or equivalently, ∂q ∂q +c − ∂t ∂x
∂q ∂α
· x,h
∂α ∂t
= ac.
(6.6.5)
Using (6.6.1), that is, replacing αt by −hxt , we obtain ∂q ∂2h ∂q +c +κ = ac, ∂t ∂x ∂x∂t
(6.6.6)
306
6 Kinematic Waves and Real-World Nonlinear Problems
∂q where κ = ( ∂α )x,h is called the diffusion coefficient. Substituting ht from (6.6.2) gives the equation for q(x, t) in the form 2 ∂a ∂q ∂ q ∂q − = ac. (6.6.7) +c −κ ∂t ∂x ∂x2 ∂x
In terms of perturbations q1 and a1 , we write q = q0 (x) + q1 (x, t) and a = 0 = a0 , we a0 (x) + a1 (x, t) where q0 and a0 are the steady-state values. Since dq dx obtain the equation 2 ∂q1 ∂ q1 ∂q1 ∂a1 = a1 . +c −κ − (6.6.8) ∂t ∂x ∂x2 ∂x This is the equation for the finite perturbation q1 (x, t) produced by a change a1 (x, t) in the rate of accumulation. When κ = 0, the equation represents the equation for q1 (x, t). In view of the difficulties in collecting data on flow curves for glaciers, we consider the shearing motion of the two-dimensional steady flow down a constant slope. We suppose that u(y) is the velocity of the layer at a height y from the ground, and τ (y) is the shearing stress. The appropriate stress–strain relation for ice is assumed as du (6.6.9) μ = τ n, dy where n ≈ 3 or 4. Further, the appropriate law for ice slipping in its bed is νu(0) = τ m (0),
(6.6.10)
where m = + 1) ≈ 2 when n = 3, and m ≈ when n = 4. If α is the angle of the slope and ρ is the density of ice, the shearing stress satisfies the equation 1 2 (n
5 2
dτ = −gρ sin α dy
(6.6.11)
with the surface condition τ (y = h) = 0. Thus, the solution of equation (6.6.11) is τ (y) = (h − y)gρ sin α.
(6.6.12)
Using this value of τ (y), we next solve the equation (6.6.9) with (6.6.10) to obtain u(y) =
1 (qρ sin α)n n+1 1 (gρ sin α)m hm + · h − (h − y)n+1 . ν μ (n + 1)
The steady flow qs per unit breadth is given by h 1 (qρ sin α)n hn+2 . qs (h) = u(y) dy = (gρ sin α)m hm+1 + ν (n + 2) 0
(6.6.13)
(6.6.14)
It is convenient to take qs (h) = A hN where A is a constant and N is approximately in the range 3 ≤ N ≤ 5. Thus, it turns out that the wave propagation velocity is dqs qs (6.6.15) = N us , =N c= dh h where us = qhs is the average steady flow velocity. This shows that the kinematic waves travel about three to five times faster than the average steady flow speed.
6.7 Roll Waves and Their Stability Analysis
307
Fig. 6.13 Flow configuration in an inclined channel.
6.7 Roll Waves and Their Stability Analysis We consider an incompressible viscous fluid flowing down an inclined channel that makes an angle α with the horizontal under the action of gravitational field g. We consider only two-dimensional unsteady flow described by the velocity field (u, v). Figure 6.13 exhibits the primary flow configuration. The unsteady two-dimensional flow is governed by the Navier–Stokes equations and the continuity equation 1 ut + uux + vuy = − px + g sin α + ν∇2 u, ρ 1 vt + uvx + vvy = − py − g cos α + ν∇2 v, ρ ux + vy = 0,
(6.7.1) (6.7.2) (6.7.3)
where ρ and ν are the density and the kinematic viscosity of the fluid. With the free surface elevation y = h(x, t), the kinematic free surface condition is given by (6.7.4) ht + uhx − v = 0 on y = h(x, t). The dynamic free surface conditions described by the tangential stress and the normal stress with pressure on the free surface are given by σxy = μ(uy + vx ) = 0 on y = h(x, t), p − + 2 ν vy + g h cos α = 0 on y = h(x, t). ρ
(6.7.5) (6.7.6)
The bottom boundary conditions are u=v=0
on y = 0.
(6.7.7)
It can be shown that the primary flow field satisfying (6.7.1)–(6.7.7) exists and is given by y y , (6.7.8a) u = U (y) = U0 2 − h h v = 0, (6.7.8b)
308
6 Kinematic Waves and Real-World Nonlinear Problems
u = p0 (y) = gρ(h − y) cos α,
h = h0 ,
(6.7.9)
where
1 2 gh sin α . (6.7.10) 2ν Based on the hydraulic approximation (that is, v is small), we integrate the continuity equation (6.7.3) with respect to y from y = 0 to y = h(x, t) and, then, combine the result with (6.7.4) to derive h(x,t) ux dx + uhx + ht = 0. (6.7.11) U0 =
0
Or equivalently, ht +
∂ ∂x
h(x,t)
u(x, y, t) dy = 0.
(6.7.12)
0
Similarly, integrating (6.7.1) with respect to y from y = 0 to y = h and using (6.7.3), (6.7.4), and (6.7.7) gives h h h ∂ ∂ 1 ∂ 2 u dy + u dy = − p dy + gh sin α − F, (6.7.13) ∂t 0 ∂x 0 ρ ∂x 0 where F represents the frictional force term given by 1 F = ν uy (x, 0, t) − uy (x, h, t) + vx (x, h, t) − p(x, h, t)hx . ρ
(6.7.14)
In view of the hydraulic approximation, the velocity component v is small, and hence, equation (6.7.2) has the approximate form −
1 ∂p − g cos α = 0. ρ ∂y
(6.7.15)
This can be integrated from y = 0 to y = h to obtain p(x, y, t) = −gρ(y − h) cos α, where p(x, h, t) = 0. Consequently, (6.7.13) reduces to the form h h ∂ 2 ∂ 1 ∂ h + gh sin α − F. u dy + u2 dy = − g cos α ∂t 0 ∂x 0 2 ∂x Using the primary flow field (6.7.9), the flow rate q is given by h 1 3 q= gh sin α. u dy = 3ν 0
(6.7.16)
(6.7.17)
(6.7.18)
We define U as the average value of u so that q = U h, and therefore, U=
1 (g sin α)h2 . 3ν
(6.7.19)
6.7 Roll Waves and Their Stability Analysis
309
Fig. 6.14 Roll waves in a steep channel.
We next write the integral involved in the second term of the left-hand side of (6.7.17) as h u2 dy = aU h = aq, (6.7.20) 0
where a is a constant parameter. Using the primary flow field (6.7.8a) in (6.7.20) gives a = 65 ≈ 1. The primary steady flow is simply caused by a balance between F and gh sin α, that is, F = gh sin α which gives, by (6.7.19), F = 3νU h−1 .
(6.7.21)
In view of (6.7.18), (6.7.20), and (6.7.21), equations (6.7.12) and (6.7.17) reduce to ∂h ∂q + = 0, ∂t ∂x ∂q 1 ∂ 2 3νU ∂q +a + g cos α h = gh sin α − , ∂t ∂x 2 ∂x h
(6.7.22) (6.7.23)
where q = hU . These two equations are known as the roll wave equations in hy1 (gh2 sin α), equadraulics. Neglecting the left-hand side of (6.7.23), so that U = 3ν tion (6.7.22) becomes ∂h ∂h + U (h) = 0. (6.7.24) ∂t ∂x It follows from this analysis that roll waves can be described by the kinematic wave equation (6.7.24), so that these waves propagate with the velocity U (h). A famous example of such waves is shown in Figure 6.14, and a beautiful photograph of this phenomenon (see Figure 6.15) was taken from a book by Cornish (1934). Cornish described his observations of roll waves with numerical data and several interesting photographs of roll waves in a long rectangular open conduit at Merligen which feeds water from a mountain to Lake Thun in the Alps. These photographs clearly show the exact periodic structure of the waves, although in this particular situation, the waves seem to be unusually long and shallow. The photograph in Figure 6.15 confirms the fact that these waves exhibit a periodic structure of a series of discontinuous bores separated by smooth profiles. Such waves frequently occur in sufficiently steep channels, for example, spillways in dams or in open channels, such as that of Figure 6.14. Observational data reported by Cornish also reveals that roll waves travel faster than the mean flow velocity.
310
6 Kinematic Waves and Real-World Nonlinear Problems
Fig. 6.15 Roll waves, looking downstream (The Grünbach, Switzerland). From Stoker (1957).
The Froude number F = (U 2 /gh) associated with this phenomenon always exceeds the critical value 2. Subsequently, Dressler (1949) made an important contribution to the subject of roll waves, and investigated how to construct nonlinear solutions of roll wave equations with appropriate jump conditions to describe the roll wave patterns. He has shown that smooth solutions can be pieced together through discontinuous bores so that the conditions describing the continuity of mass and momentum across the discontinuity are satisfied. For more details, the reader is referred to Dressler’s paper (1949). We close this section by adding the stability analysis of roll waves. We introduce small perturbations to the primary flow field as (6.7.25) h = h0 + η exp ik(x − ct) , U = U 0 + u exp ik(x − ct) , (6.7.26) where
2 gh20 sin α = U0 . (6.7.27) 3ν 3 Substituting (6.7.25), (6.7.26) into (6.7.22) and (6.7.23) and retaining only terms linear in η and u leads to the following equations: U 0 − c η + h0 u = 0, (6.7.28) 2 −ikc U 0 η + h0 u + ika U 0 η + 2U 0 h0 u + ikg h0 η cos α 3ν 3νU 0 = gη sin α − u+ η. (6.7.29) h0 h20 U0 =
6.8 Simple Waves and Riemann’s Invariants
311
Fig. 6.16 A typical continuous roll wave solution.
We next use (6.7.28) to express u in terms of η, so that equation (6.7.29) can be written as 2 ik −c U 0 − U 0 − c + a U 0 − 2U 0 U 0 − c + gh0 cos α 3ν (6.7.30) = g sin α + 2 U 0 − c + U 0 . h0 Or equivalently, ik −c2 + a 2c − U 0 U 0 + gh0 cos α = 3νh−2 3U 0 − c . 0
(6.7.31)
The critical state exists for values of c which are given by c = 3U 0 = 2U0 and c2 − a 2c − U 0 U 0 − (g h0 ) cos α = 0.
(6.7.32a) (6.7.32b)
The former gives ( Uc0 ) = 2, and this result can be used to express (6.7.32b) in terms of the critical Froude number Fc = (U02 /gh0 ) as 5a (6.7.33) Fc = cos α 4 1− 9 for various values of a. In particular, when a = 27 25 , the growth rate of the instability can be obtained as 5 (6.7.34) F > Fc = cos α. 8 This result can also be derived from the stability analysis of the original equations (6.7.1)–(6.7.3). In addition to the preceding discontinuous roll wave solutions, we conclude this section by mentioning Dressler’s continuous roll wave solutions which describe periodic, progressive waves flowing over a sloping bottom in terms of cnoidal waves, as shown in Figure 6.16. For further mathematical details, the reader is referred to Dressler’s (1949) paper.
6.8 Simple Waves and Riemann’s Invariants In one of his famous papers, Riemann (1858) made a remarkable discovery which laid the foundation for all subsequent work on the theory and applications of nonlinear plane sound waves and simple waves. Based on his general formulation, we
312
6 Kinematic Waves and Real-World Nonlinear Problems
discuss the mathematical analysis of the theory of simple waves and Riemann invariants. To develop the governing equations for finite-amplitude wave motions in gas dynamics, we consider the continuity equation and momentum equations of an unsteady compressible fluid without body forces in the form Dρ + ρ(∇ · u) = 0, Dt Du = −∇p, ρ Dt
(6.8.1) (6.8.2)
where ρ is the density, u = (u, v, w) is the fluid velocity vector, p is the pressure, and the total derivative is D ∂ ≡ + u · ∇. (6.8.3) Dt ∂t It follows from thermodynamics that density is a function of pressure and entropy, that is, ρ = ρ(p, s), where s is the entropy. Hence, ∂ρ ∂ρ dρ = dp + ds. (6.8.4) ∂p s ∂s p For isentropic flow, s is constant, that is, ds = 0. Thus, equation (6.8.4), written in terms of the total derivative following a fluid element, becomes 1 Dp Dρ = 2 , Dt c Dt
(6.8.5)
where c is the velocity of sound, so that c2 =
dp . dρ
(6.8.6)
Substituting (6.8.5) into (6.8.1) gives 1 Dp + ρ∇ · u = 0. c2 Dt
(6.8.7)
For one-dimensional unsteady isentropic flow, equations (6.8.7) and (6.8.2) become ∂u ∂p 1 ∂p +ρ +u = 0, (6.8.8) 2 c ∂t ∂x ∂x ∂u ∂p ∂u ρ + +u = 0. (6.8.9) ∂t ∂x ∂x Adding (6.8.8) and (6.8.9) gives
1 ∂p ∂u ∂p ∂u + = 0. + (u + c) + (u + c) ∂t ∂x ρc ∂t ∂x
(6.8.10)
6.8 Simple Waves and Riemann’s Invariants
Subtracting (6.8.8) from (6.8.9) yields
∂u 1 ∂p ∂p ∂u + (u − c) − + (u − c) = 0. ∂t ∂x ρc ∂t ∂x
313
(6.8.11)
In general and in principle, a solution of these equations gives u = u(x, t) and p , we rewrite (6.8.10) and (6.8.11) to obtain p = p(x, t). However, introducing P = ρc ∂ ∂ (u + P ) + (u + c) (u + P ) = 0, ∂t ∂x ∂ ∂ (u − P ) + (u − c) (u − P ) = 0. ∂t ∂x
(6.8.12) (6.8.13)
According to the theory of characteristics for first-order, quasi-linear equations, equations (6.8.12), (6.8.13) signify that u + P = const. u − P = const.
dx = u + c, dt dx along a curve C− such that = u − c. dt along a curve C+ such that
(6.8.14) (6.8.15)
These are truly remarkable results first obtained by Riemann (1858). Physically, the curves C+ and C− in the (x, t)-plane are referred to as positive and negative wavelets, respectively. In general, these wavelets do not travel at constant speed, but, everywhere, they move with the local speed of sound relative to the fluid velocity. More precisely, a curve C+ is the locus of a point (x, t) which always moves forward (that is, in the positive x-direction) at the local wave speed c relative to the local fluid velocity u. Similarly, a curve C− is the locus of a point in the (x, t)-plane which always moves backward, that is, in the negative x-direction. To gain further physical understanding of the Riemann theory, we examine an initial-value problem. If, at time t = 0, all the fluid except that in a certain slab of the x-axis is undisturbed, that is, u = 0 and P = 0 at t = 0, while the disturbance within the slab may be large but, of course, with constant entropy. The main question is how a disturbance propagates ahead of and behind the slab in the subsequent motion. We draw Figure 6.17 in the (x, t)-plane. At t = 0, the disturbance is confined to the slab between the points B (for back) and F (for front). To investigate the subsequent development of the flow, we consider a large number of C+ curves shown on the figure B F including those designated as C+ and C+ , which originate at t = 0 from the points B and F , and, similarly, with the C− curves. The slope of these curves in not known in advance because it depends on how u and c change. However, some important information about them is described below. First, equation (6.8.15) indicates that u+P takes a constant value along each C− curve, but different for different members of F and all curves ahead of it originate from the region the C− family. However, C− ahead of F , where u = P = 0. The constant value of u − P on each of these can F only be zero. Thus, we conclude that u = P ahead of C− . Throughout this region F ahead of C− , which increases progressively wider than the region ahead of the slab,
314
6 Kinematic Waves and Real-World Nonlinear Problems
Fig. 6.17 Riemann’s initial-value problem.
we examine how disturbances propagate. Similar arguments reveal that u = P beB B hind C− . Analogous arguments for the C+ curves indicate that u = −P behind C+ F F and also ahead of C+ . It follows from the above analysis that C+ must lie ahead of F F C− , and the region ahead of C+ remains undisturbed with u = P = 0 because there B are no disturbances in this region. Similar comments apply to the region behind C− , that is, this region is also undisturbed with u = P = 0. Figure 6.17 also indicates F B F B intersects C+ at a point A and the region C− AC+ that, after some time t = td , C− remains undisturbed with u = P = 0. Evidently, during the time interval 0 < t < td , the disturbances become disentangled, and thereafter, they propagate as two simple waves—one forward and one backward with an undisturbed region in between. So, the upshot of the Riemann analysis is that the disturbance initially confined to BF becomes disentangled when t = td into a pair of simple waves propagating in opposite directions (one traveling to the right and the other traveling to the left), separated by an undisturbed region. To give a more general formulation of the propagation of simple waves, we rewrite the equations (6.8.8), (6.8.9) for one-dimensional, unsteady, isentropic flow in terms of density ρ as ∂ρ ∂u ∂ρ +u +ρ = 0, ∂t ∂x ∂x ∂u ∂u c2 ∂ρ = 0. +u + ∂t ∂x ρ ∂x
(6.8.16) (6.8.17)
In matrix form, this system of equations is A
∂U ∂U +I = 0, ∂x ∂t
where U , A, and I are matrices given by
(6.8.18)
6.8 Simple Waves and Riemann’s Invariants
U=
ρ , u
A=
u ρ c2 /ρ u
,
and
I=
1 0 0 1
315
.
(6.8.19)
It is of interest to determine how a solution evolves with time t. Hence, we leave the time variable unchanged and replace the space variable x by some arbitrary curvilinear coordinate ξ so that the semicurvilinear coordinate transformation from (x, t) to (ξ, τ ) can be introduced by ξ = ξ(x, t),
τ = t.
(6.8.20)
If the Jacobian of this transformation is nonzero, we can transform (6.8.19) by the following correspondence rule: ∂ξ ∂ ∂τ ∂ ∂ξ ∂ ∂ ∂ ≡ + · = + , ∂t ∂t ∂ξ ∂t ∂τ ∂t ∂ξ ∂τ (6.8.21ab) ∂ ∂ξ ∂ ∂τ ∂ ∂ξ ∂ ≡ + · = . ∂x ∂x ∂ξ ∂x ∂τ ∂x ∂ξ This rule transforms equation (6.8.18) into the form ∂U ∂ξ ∂ξ ∂U I + I+ A = 0. ∂τ ∂t ∂x ∂ξ
(6.8.22)
This equation can be used to determine ∂U/∂ξ provided that the determinant of its coefficient matrix is nonzero. Obviously, this condition depends on the nature of the curvilinear coordinate curves ξ(x, t) = const., which has been kept arbitrary. We assume now that the determinant vanishes for the particular choice ξ = η, so that ∂η I + ∂η A = 0. (6.8.23) ∂t ∂x In view of this, ∂U/∂η will become indeterminate on the family of curves η = const., and hence, ∂U/∂η may be discontinuous across any of the curves η = const. This implies that each element of ∂U/∂η will be discontinuous across any of the curves η = const. Then, it is necessary to find how these discontinuities in the elements of ∂U/∂η are related across the curve η = const. We next consider the solutions U which are everywhere continuous with discontinuous derivatives ∂U/∂η across the particular curve η = const. = η0 . Since U is continuous, elements of the matrix A are not discontinuous across η = η0 , so that A can be determined in the neighborhood of a point P on η = η0 . And since ∂/∂τ represents differentiation along the curves η = const., ∂U/∂τ is continuous everywhere, and hence, it is continuous across the curve η = η0 at P . In view of these facts, it follows that differential equation (6.8.22) across the curve ξ = η = η0 at P gives
∂U ∂η ∂η I+ A = 0, (6.8.24) ∂t ∂x P ∂η p
316
6 Kinematic Waves and Real-World Nonlinear Problems
where [f ]p = f (P +) − f (P −) denotes the discontinuous jump in the quantity f across the curve η = η0 , and f (P −) and f (P +) represent the values to the immediate left and immediate right of the curve at P . Since P is any arbitrary point on the curve, ∂/∂η denotes the differentiation normal to the curves η = const., so that equation (6.8.24) can be regarded as the compatibility condition satisfied by ∂U/∂η on either side of and normal to these curves in the (x, t)-plane. Obviously, (6.8.24) is a homogeneous system of equations for the two jump quantities [∂U/∂η]. Therefore, for the existence of a nontrivial solution, the coefficient determinant must vanish, that is, ∂η I + ∂η A = 0. (6.8.25) ∂t ∂x However, along the curves η = const., we have dx 0 = dη = ηt + ηx , dt
(6.8.26)
so that these curves have the slope ηt dx =− =λ dt ηx
(say).
(6.8.27)
Consequently, equations (6.8.24) and (6.8.25) can be expressed in terms of λ in the form
∂U (A − λI) = 0, (6.8.28) ∂η |A − λI| = 0, (6.8.29) where λ represents the eigenvalues of the matrix A and [∂U/∂η] are proportional to the corresponding right eigenvectors of A. Since A is a 2 × 2 matrix, it must have two eigenvalues. If these are real and distinct, integration of (6.8.27) leads to two distinct families of real curves Γ1 and Γ2 in the (x, t)-plane: dx Γ1 : = λr , r = 1, 2. (6.8.30ab) dt The families of curves Γr are called the characteristic curves of the system (6.8.18). Any one of these families of curves Γr may be chosen for the curvilinear coordinate curves η = const. The eigenvalues λr have the dimensions of velocity, and the λr associated with each family will then be the velocity of propagation of the matrix column vector [∂U/∂η] along the curves Γr belonging to that family. In this particular case, the eigenvalues λ of the matrix A are determined by (6.8.29), that is, by the equation u − λ ρ 2 (6.8.31) c /ρ u − λ = 0,
6.8 Simple Waves and Riemann’s Invariants
317
so that λ = λr = u ± c,
(6.8.32ab)
r = 1, 2.
Consequently, the families of the characteristic curves Γr (r = 1, 2) defined by (6.8.30ab) become Γ1 :
dx = u + c, dt
and
Γ2 :
dx = u − c. dt
(6.8.33ab)
In physical terms, these results indicate that disturbances propagate with the sum of the velocities of the fluid and sound along the family of curves Γ1 . In the second family Γ2 , they propagate with the difference of the fluid velocity u and the sound velocity c. (1) r are solutions of the equations The right eigenvectors Rr ≡ R R(2) r
(A − λr I)Rr = 0,
(6.8.34)
r = 1, 2,
or equivalently,
u − λr c2 /ρ
ρ u − λr
(1)
Rr (2) Rr
= 0,
r = 1, 2.
This result combined with (6.8.29) gives (1) Rr [ρη ] 1 = =α , (2) ±c/ρ [Rη ] Rr
r = 1, 2,
(6.8.35ab)
(6.8.36)
where α is a constant. In other words, across a wavefront in the Γ1 family of characteristic curves, [∂ρ/∂η] [∂u/∂η] = , (6.8.37) 1 c/ρ and across a wavefront in the Γ2 family of characteristic curves, [∂u/∂η] [∂ρ/∂η] = , 1 −c/ρ
(6.8.38)
where c and ρ have values appropriate to the wavefront. To introduce the Riemann invariants explicitly, we form the linear combination of the eigenvectors (±c/ρ, 1) with equations (6.8.16), (6.8.17) to obtain c c2 (6.8.39) ± (ρt + ρux + uρx ) + ut + uux + ρx = 0. ρ ρ We use ∂u/∂ρ = ±c/ρ from (6.8.37), (6.8.38) and rewrite (6.8.39) as ±
c ρt + (u ± c)ρx + ut + (u ± c)ux = 0. ρ
(6.8.40)
318
6 Kinematic Waves and Real-World Nonlinear Problems
In view of (6.8.33ab), equation (6.8.40) becomes c du ± dρ = 0 ρ or equivalently,
on Γr , r = 1, 2,
d F (ρ) ± u = 0
where
ρ
F (ρ) = ρ0
on Γr ,
(6.8.41)
(6.8.42)
c(ρ) dρ. ρ
(6.8.43)
Integration of (6.8.42) gives F (ρ) + u = 2r
on Γ1
and
F (ρ) − u = 2s
on Γ2 ,
(6.8.44ab)
where 2r and 2s are constants of integration on Γ1 and Γ2 , respectively. The quantities r and s are called the Riemann invariants. As stated above, r is an arbitrary constant on characteristics Γ1 , and hence, in general, r will vary on each Γ2 . Similarly, s is constant on each Γ2 but will vary on Γ1 . It is natural to introduce r and s as new curvilinear coordinates. Since r is constant on Γ1 , s can be treated as the parameter on Γ1 . Similarly, r can be regarded as the parameter on Γ2 . Then, dx = (u±c) dt on Γr implies that dt dx = (u + c) ds ds dt dx = (u − c) dr dr
on Γ1 ,
(6.8.45)
on Γ2 .
(6.8.46)
In fact, r is constant on Γ1 , and s is constant on Γ2 . Therefore, the derivatives in equations (6.8.45), (6.8.46) are partial derivatives, with respect to s and r, so that we can rewrite them as ∂t ∂x = (u + c) , ∂r ∂s ∂x ∂t = (u − c) . ∂r ∂r
(6.8.47) (6.8.48)
These two first-order partial differential equations, in general, can be solved for x = x(r, s) and t = t(r, s), and then, by inversion, r and s can be obtained as functions of x and t. Once this is done, we use (6.8.44ab) to determine u(x, t) and ρ(x, t) in terms of r and s as u(x, t) = r − s, F (ρ) = r + s.
(6.8.49a) (6.8.49b)
When one of the Riemann invariants r and s is constant throughout the flow, the corresponding solution is tremendously simplified. The solutions are known as simple wave motions representing simple waves in one direction only. The generating
6.8 Simple Waves and Riemann’s Invariants
319
mechanisms of simple waves with their propagation laws can be illustrated by the piston problem in gas dynamics (see Example 6.8.2). We close this section by reducing the two first-order equations (6.8.47) and (6.8.48) to a single second-order linear hyperbolic equation. This can be done by equating xrs and xsr which eliminates x from the system (6.8.47), (6.8.48), so that we obtain the single equation (u + c)trs + (ur + cr )ts = (u − c)trs + (us − cs )tr .
(6.8.50)
The coefficients of this equation are functions of r and s, and they can be determined explicitly. First, we find that ur = 1 and us = −1 from (6.8.49a). Second, from (6.8.49b), we calculate F (ρ)ρr = F (ρ)ρs = 1, and hence, from result (6.8.43), F (ρ) = ρc so that ρr = ρs = ρc . Finally, we determine that cr = c (ρ)ρr =
ρ · c (ρ), c
cs = c (ρ)ρs =
ρ · c (ρ). c
These results are used to simplify (6.8.50) further as ρ dc ρ dc 2ctrs + 1 + ts = − 1 + tr , c dρ c dρ or equivalently,
1 ρ dc trs + 1+ (tr + ts ) = 0. 2c c dρ
(6.8.51ab)
(6.8.52)
(6.8.53)
The coefficient of (tr +ts ) depends only on ρ, and hence, by (6.8.49b), it is a function of (r + s), so that we can write it as f (r + s). More explicitly, we write ρ dc 1 1+ = f (r + s). (6.8.54) 2c c dρ Consequently, equation (6.8.53) reduces to the second-order hyperbolic equation trs + f (r + s)(tr + ts ) = 0.
(6.8.55)
Once we can solve this equation for t = t(r, s), it can be used to solve the system (6.8.47), (6.8.48) for x = x(r, s). Example 6.8.1 Riemann’s Invariants for a Polytropic Gas. A polytropic gas is usually characterized by the pressure–density law p = kργ where k and γ are constants. In this case, the velocity of sound c is calculated by using the formula c2 = kγργ−1 . It follows from this result that ct 2 2ρ c2 ccx = ρx . and (6.8.56) ρt = γ−1 c r−1 ρ In view of these results, equations of motion (6.8.16), (6.8.17) assume the form
320
6 Kinematic Waves and Real-World Nonlinear Problems
γ−1 cux = 0, 2 2 ccx = 0. ut + uux + γ−1
ct + ucx +
(6.8.57) (6.8.58)
We divide (6.8.57) by (γ − 1) and (6.8.58) by 2, and then add them and subtract the former from the latter to obtain the following equations (Riemann 1858): ∂r ∂r + (c + u) = 0, ∂x ∂x
∂s ∂s − (c − u) = 0, ∂t ∂x
(6.8.59ab)
where r and s are the Riemann invariants given by 1 c + u, γ−1 2 c 1 s= − u. γ−1 2
r=
(6.8.60a) (6.8.60b)
Equations (6.8.60a), (6.8.60b) signify that 1 c + u = const. γ−1 2 1 c − u = const. s= γ−1 2
r=
dx = c + u, dt dx on = c − u. dt on
(6.8.61) (6.8.62)
Furthermore, u = r − s = const., and c = 12 (γ − 1)(r + s) = const. confirming the fact that u and c are separately constant. Consequently, the characteristics dx dt = c ± u = const., which represents two families of straight lines in the (x, t)-plane. Mathematically, the characteristics exist because the governing quasilinear equations (6.8.57) and (6.8.58) are hyperbolic. Physically, they represent the wavelets propagating in both directions with speed c while also connected by the fluid with local speed u. If, as often happens, one of the Riemann invariants is constant everywhere (that is, it takes the same constant value on every characteristic line of its family), the resulting solution is said to form a simple wave. This example simply illustrates Riemann’s original general approach to the problem. However, it should be pointed out that the Riemann invariant can be deriveddirectly from (6.8.44ab) by using (6.8.43) combined with the sound speed √ 1 dp = kγρ 2 (γ−1) . c = dρ The solution of the equations of motion in the above simple wave assumes the form u(x, t) = f (ξ), ξ = x − (u + c)t, (6.8.63) where u(x, 0) = f (x) represents the initial data. If f (x) ≥ 0, u + c will be everywhere positive, and hence, the family of characteristics carrying the simple wave has positive slope in the (x, t)-plane, and the wave is said to be forward progressing. The speed of propagation of the forward progressing wave is given by
6.8 Simple Waves and Riemann’s Invariants
u+c=
1 (γ + 1)u + c0 , 2
321
(6.8.64)
where a slightly different form of (6.8.61), that is, c = 12 (γ + 1)u + const., is used, and c0 is the sound speed of a one-dimensional compressive wave which cannot remain continuous indefinitely, whereas a similar expansion wave can. Differentiating (6.8.63) combined with (6.8.64) gives f (ξ) ∂u = f (ξ)ξx = , ∂x 1 + 12 (γ + 1)t f (ξ)
1 f (ξ)f (ξ) ∂u = f (ξ)ξt = − (γ + 1) . ∂t 2 1 + 12 (γ + 1)t f (ξ)
(6.8.65) (6.8.66)
Evidently, both ux and ut tend to infinity as t→−
2 , (γ + 1)f (ξ)
(6.8.67)
which is positive provided that f (ξ) < 0, corresponding to the compressive part of the wave profile. The discontinuity occurs first at the instant t=
2 . (γ + 1){−f (ξ)}max
(6.8.68)
The solution develops a discontinuity in the form of shock waves. The gas is at rest. From the constant value in this simple wave of the Riemann invariant, it is evident that the greater the speed u, the greater the value of c and, therefore, of the pressure p. Hence the largest velocities and pressures move most rapidly through the gas. The peaks of velocity or pressure tend to overtake the troughs, thus, leading to the breakdown of the uniqueness of the solution whenever the initial profile contains a compressive part. This has already been illustrated in Chapter 5 (see Figure 5.4) for a similar situation observed in the case of a general, nonlinear initial-value problem. As time increases, the initial velocity (or pressure) profile is progressively distorted, and the part of the wave profile that was initially expansive (velocity increases in the direction of motion) tends to elongate. Ultimately, the compressive part of the wave profile develops multiple-valuedness and then breaks at the point when it becomes vertical. This simple analysis confirms that, in inviscid flow, a shock wave develops in one-dimensional unsteady flow as exhibited in the photographs included in Figure 6.18, which is due to Dr. W. Bleakney and Dr. Wayland Griffith at the Palmer Physical Laboratory, Princeton University. This photograph shows that in (a) the density gradient is moderate, in (b) it becomes steeper, and in (c) it is clearly discontinuous. The steepening effect of the waveform for the case of one-dimensional flow is equally true for flows with cylindrical and spherical symmetry. This fundamental result was obtained long ago by Burton (1893) by intuitive arguments and has been confirmed by Pack (1960) by the method of characteristics. We complete this example by adding Riemann’s original approach to an explicit solution for the problem of polytropic gas. First, we can calculate the unknown function f (r + s) in equation (6.8.55) to find the explicit form of this equation. First, we √ 1 use c = kγρ 2 (γ−1) to obtain c (ρ) and then (6.8.60a), (6.8.60b) to find F (ρ)
322
6 Kinematic Waves and Real-World Nonlinear Problems
Fig. 6.18 Development of shock waves at successive instants. From Lighthill (1956).
c (ρ) =
c 1 (γ − 1) 2 ρ
and
F (ρ) =
2c . (γ − 1)
(6.8.69ab)
Thus, it follows from (6.8.54) that f (r + s) =
1 1 1 γ+1 n (γ + 1) = = , 4c 2 γ − 1 (r + s) (r + s)
(6.8.70)
where n = 12 ( γ+1 γ−1 ) = const. Consequently, equation (6.8.55) reduces to the explicit form n (tr + ts ) = 0, (6.8.71) trs + r+s where r and s are involved in this equation symmetrically. This allows us to extend the method to (6.8.71) by introducing a new dependent variable w(r, s) so that t(r, s) = g(r + s)w(r, s),
(6.8.72)
where g is to be determined so that wr and ws vanish. We substitute (6.8.72) into (6.8.71) to derive ng 2n (wr + ws ) + g 1 + w = 0. (6.8.73) g wrs + g + r+s r+s To simplify the problem further, we assume that g(r + s) = (r + s)−n , so that (6.8.73) reduces to the canonical hyperbolic equation wrs +
n(1 − n) w = 0. (r + s)2
(6.8.74)
Indeed, this is a telegraph equation with a variable coefficient. Following Riemann analysis (1858), we introduce the Riemann function R(r, s, ρ, σ), which must be a function of (r, s), and look for the Riemann function as a function of a single variable z=
(r − ρ)(s − σ) . (r + s)(ρ + σ)
(6.8.75)
More explicitly, we set R(r, s, ρ, σ) = g(z) so that R satisfies (6.8.74), and hence, g(z) can be determined. We also assume that R satisfies some additional conditions
6.8 Simple Waves and Riemann’s Invariants
∂R = 0 for r = ρ, ∂s
∂R =0 ∂r
323
for s = σ,
and R(ρ, σ, ρ, σ) = 1. (6.8.76) We next substitute R(r, s, ρ, σ) = g(z) in (6.8.74) to generate the ordinary differential equation for g(z) in the form z(1 − z)g (z) + (1 − 2z)g (z) − n(1 − n)g(z) = 0.
(6.8.77)
This is the hypergeometric equation z(1 − z)g + c − (1 + a + b)z g − abg = 0
(6.8.78)
with a = 1 − n, b = n, and c = 1. The first two conditions in (6.8.76) are automatically satisfied by R(r, s, ρ, σ) = g(z), and the last condition in (6.8.76) needs g(0) = 1. The solution of (6.8.78) is the standard hypergeometric function R(r, s, ρ, σ) = F (1 − n, n, 1, −z), where F (a, b, c, z) is the hypergeometric function defined by ab a(a + 1)b(b + 1) z 2 F (a, b, c, z) = 1 + z+ + ··· . c c(c + 1) 2!
(6.8.79)
(6.8.80)
For the special case in which n is an integer, a transformation τ = (r + s)−1 (tr + ts ) can be used to solve the original equation (6.8.71) which becomes n+1 (τr + τs ) = 0. τrs + r+s
(6.8.81)
(6.8.82)
This shows that, if a solution of (6.8.71) exists for any n, then the use of the transformation (6.8.81) leads us to a solution for n + 1. In particular, when n = 0, equation (6.8.71) reduces to the canonical hyperbolic equation trs = 0,
(6.8.83)
which gives a solution by direct integration as t(r, s) = φ(r) + ψ(s),
(6.8.84)
where φ and ψ are arbitrary functions. Using (6.8.81), the general solution of (6.8.82) with n = 0, that is, of (6.8.71) with n = 1, is given by τ (r, s) = Similarly, for n = 2, we obtain
φ (r) + ψ (s) . (r + s)
(6.8.85)
324
6 Kinematic Waves and Real-World Nonlinear Problems
τ (r, s) = (r + s)−1 φ (r) + ψ (s) − 2(r + s)−2 φ (r) + ψ (s) . In general, when n is an integer, the solution of (6.8.71) is given by
φ(r) ∂ n−1 ψ(s) ∂ n−1 + , t(r, s) = n−1 ∂r (r + s)n ∂sn−1 (r + s)n
(6.8.86)
(6.8.87)
where φ(r) and ψ(s) are arbitrary functions. This result is also valid for fractional values of n and can be expressed in terms of fractional derivatives with n = m + α, m a nonnegative integer, and 0 ≤ α < 1. Consequently, (6.8.87) takes the form
φ(r) ψ(s) ∂m ∂m −(1−α) t(r, s) = m · 0 Dr −(1−α) + , · D 0 s ∂r (r + s)m+α ∂sm (r + s)m+α (6.8.88) where the fractional integrals (Debnath 1995) are defined by r 1 −(1−α) D f (r) = (r − x)−α f (x) dx. 0 r Γ (1 + α) 0
(6.8.89)
Result (6.8.88) is valid for n ≥ 0 and for n < 0. Then (6.8.88) also holds if m is set equal to zero and α = n. We consider another special case in which n is half-integer. It is convenient to handle this case with u and c instead of r and s as independent variables, where u=r−s
and c =
γ−1 (r + s). 2
(6.8.90ab)
Using these results, we change the variables from r, s to u, c so that equation (6.8.71) reduces to 2 γ−1 2n tcc + (6.8.91) tuu − tc = 0. 2 c This is known as the Euler–Poisson–Darboux equation. If m is an integer and 2n = m − 1, then equation (6.8.91) becomes tuu −
γ−1 2
2 tcc +
m−1 tc c
= 0.
(6.8.92)
This can be treated as a wave equation in m space variables for a problem with spherical symmetry. To check this situation, we take c = x21 + x22 + · · · + x2m , and write ∂2 ∂2 ∂2 + + · · · + , (6.8.93) ∇2 ≡ ∂x21 ∂x22 ∂x2m and then assume that t = t(x1 , x2 , . . . , xm , u) = t(c, u). With these assumptions, we obtain m−1 2 tc . (6.8.94) ∇ t = tcc + c
6.8 Simple Waves and Riemann’s Invariants
325
Fig. 6.19 Simple waves generated by the motion of a piston.
Then, equation (6.8.92) reduces to the form tuu −
γ−1 2
2 ∇2 t = 0.
(6.8.95)
This is a wave equation where the waves travel with the speed 12 (γ − 1). Example 6.8.2 The Piston Problem in a Polytropic Gas. The problem is to determine how a simple wave is produced by the prescribed motion of a piston in the closed end of a semi-infinite tube filled with gas. This is a one-dimensional unsteady problem in gas dynamics. We assume that the gas is initially at rest with a uniform state u = 0, ρ = ρ0 , and c = c0 . The piston starts from rest at the origin and is allowed to withdraw from the tube with a variable velocity for a time t1 , after which the velocity of withdrawal remains constant. The piston path is shown by a dotted curve in Figure 6.19. In the (x, t)-plane, the path of the piston is given by x = X(t) with X(0) = 0. The fluid velocity u is equal to the ˙ piston velocity X(t) on the piston x = X(t), which will be used as the boundary condition on the piston. The initial state of the gas is given by u = u0 , ρ = ρ0 , and c = c0 at t = 0 for x ≥ 0. The characteristic line Γ0 that bounds it and passes through the origin is determined by the equation dx = (u + c)t=0 = c0 , dt so that the equation of the characteristic line Γ0 is x = c0 t. In view of the uniform initial state, all of the Γ2 characteristics start on the x-axis, so that the Riemann invariants s in (6.8.60b) must be constant and have the form 2c0 2c −u= , γ−1 γ−1 or equivalently,
(6.8.96)
326
6 Kinematic Waves and Real-World Nonlinear Problems
2(c − c0 ) , γ−1 γ−1 c = c0 + u. 2
(6.8.97a)
u=
(6.8.97b)
The characteristics Γ1 meeting the piston are given by 2c + u = 2r γ−1
on each Γ1 :
dx = u + c, dt
(6.8.98)
which is, by (6.8.97a), (6.8.97b) which holds everywhere, u = const.
on Γ1 :
dx 1 = c0 + (γ + 1)u. dt 2
(6.8.99)
Since the flow is continuous with no shocks, u = 0 and c = c0 ahead of and on Γ0 , which separates those Γ1 meeting the x-axis from those meeting the piston. The family of lines Γ1 through the origin has the equation (dx/dt) = ξ, where ξ is a parameter with ξ = c0 on Γ0 . The Γ1 characteristics are also defined by (dx/dt) = u + c, so that ξ = u + c. Hence, eliminating c from (6.8.97b) gives 2 u= (ξ − c0 ). (6.8.100) γ−1 Substituting this value of u in (6.8.97b), we obtain 2c0 γ−1 ξ+ . c= γ−1 γ−1
(6.8.101)
If follows from c2 = γkργ−1 and (6.8.97b) with the initial data ρ = ρ0 and c = c0 that
2/(γ−1) γ−1 ρ = ρ0 1 + u . (6.8.102) 2c0 With ξ = x/t, results (6.8.100) through (6.8.102) give the complete solution of the piston problem in terms of x and t. Finally, the equation of the characteristic line Γ1 is found by integrating the second equation of (6.8.99) and using the boundary condition on the piston. When a line ˙ ) along it, and the equation Γ1 intersects the piston path at time t = τ , then u = X(τ becomes γ+1 ˙ x = X(τ ) + c0 + X(τ ) (t − τ ). (6.8.103) 2 Note that the family Γ1 represents straight lines with slope (dx/dt) increasing with velocity u. Consequently, the characteristics are likely to overlap on the piston, ˙ ) > 0 for any τ . If u increases, so do c, ρ, and p, so that instability that is, X(τ develops. This shows that shocks will be formed in the compressive part of the disturbance.
6.8 Simple Waves and Riemann’s Invariants
327
Fig. 6.20 Long surface waves in shallow water.
Example 6.8.3 Shallow Water Waves. We consider the propagation of surface waves whose wavelength is large compared with the depth h(x), as shown in Figure 6.20. In terms of the horizontal velocity component u = u(x, t) and the free surface elevation η = η(x, t), shallow water equations are given by
ut + uux = −gηx , u(η + h) x = −ηt ,
(6.8.104) (6.8.105)
where g is the acceleration due to gravity. Introducing the propagation speed c = g g(η + h), so that cx = 2c (ηx + hx ) and ct = (gηt /2c), we reformulate the preceding equations in terms of c to obtain ut + uux + 2ccx − Hx = 0, 2(ct + u cx ) + c ux = 0,
(6.8.106) (6.8.107)
where H = g h(x). For the particular case in which Hx = m = const., the slope of the bottom is constant. Adding (6.8.106) and (6.8.107) leads to the equation
∂ ∂ (u + 2c − mt) = 0. (6.8.108) + (u + c) ∂t ∂x Similarly, subtracting (6.8.107) from (6.8.106) gives
∂ ∂ (u − 2c − mt) = 0. + (u − c) ∂t ∂x
(6.8.109)
This system of equations (6.8.108) and (6.8.109) is very similar to that of the equations for one-dimensional sound waves (6.8.12) and (6.8.13). So, we may follow the same analysis to investigate this problem. Equations (6.8.106), (6.8.107) state that the functions (u ± 2c − mt) are constant on the two sets of characteristics C± : dx dt = u ± c, respectively. It may also be observed that the two families of characteristics C± are really distinct because c = g(η + h) = 0, since we assume that η > −h, that is, the free surface never coincides with the bottom. In particular, for a fluid of constant depth h, Hx = 0, and the quantity m in (6.8.108) and (6.8.109) is zero. Thus the functions (u ± 2c) are constants on the
328
6 Kinematic Waves and Real-World Nonlinear Problems
two families of characteristics C± : dx dt = u ± c. Further, one of the two families of characteristics C± consists of straight lines along each of which u and c are constant. Consequently, the corresponding motion can be referred to as a simple wave. As in the case of the initial-value problem for sound waves treated at the beginning of this Section 6.8, the shallow water wave problem can easily be handled. The only difference is that, in the present situation, the solution will develop what is called a bore or hydraulic jump. In other words, the compressive part of the water waves will always break, that is, the development of a bore or hydraulic jump in water is very similar to the development of shock waves in gas dynamics. Example 6.8.4 Nonlinear Wave Equation. We consider the nonlinear wave equation in the form (6.8.110) utt − c2 (ux )uxx = 0, where, as indicated, the velocity c depends on the slope of u. We transform this equation into a system of two equations by introducing ut = v, and ux = w, so that (6.8.110) reduces to the system of coupled equations vx − wt = 0, vt − c2 (w)wx = 0.
(6.8.111) (6.8.112)
We next rearrange the terms of the second equation to write the system in the matrix form 0 −1 vt 1 0 vx + = 0. (6.8.113) 1 0 0 −c2 (w) wx wt According to (6.8.29), the eigenvalue equation for λ is 1 −λ = 0. |A − λI| = λ −c2
(6.8.114)
So, the eigenvalues λ are λ = ±c,
(6.8.115)
and the corresponding eigenvectors are (c ∓ 1). Consistent with the notation used before, we obtain two families of characteristics as C+ : λ =
dx = c; dt
C− : λ =
dx = −c. dt
(6.8.116)
We multiply (6.8.111) by c and then add the result with (6.8.112) to obtain (vt + cvx ) − c(wt + cwx ) = 0.
(6.8.117)
Similarly, we multiply (6.8.111) by c and subtract the result from (6.8.112) to get (vt − cvx ) + c(wt − cwx ) = 0. We replace ±c by
dx dt
in (6.8.117) and (6.8.118) to find
(6.8.118)
6.9 The Nonlinear Hyperbolic System and Riemann’s Invariants
dv ± c(w) dw = 0 on C± . We then write c(w) dw = dF (w) in the last equation, so that d v ± F (w) = 0 on C± .
329
(6.8.119)
(6.8.120)
This enables us to introduce Riemann invariants r and s as before by 2s = v − F (w).
2r = u + F (w),
(6.8.121ab)
We can now continue the same mathematical analysis following equations (6.8.44ab), and so the rest is left to the reader as an exercise.
6.9 The Nonlinear Hyperbolic System and Riemann’s Invariants One of the simplest nonlinear (or quasi-linear) hyperbolic systems of evolution equations describing wave propagation has the form ∂U ∂U + A(U ) = 0, ∂t ∂x where
U=
u1 u2
and
A=
a11 a21
a12 a22
(6.9.1)
with aij = aij (u1 , u2 ).
In this system, the elements aij of the matrix A depend explicitly on u1 and u2 , but only implicitly on x and t through u1 and u2 . Following the discussion of Section 6.8, the real eigenvalues are the solutions of the equation |A − λI| = 0. (6.9.2) The system will be called strictly hyperbolic if the eigenvalues λ1 and λ2 are real and distinct. We denote the left eigenvectors L1 and L2 corresponding to the eigenvalues λ1 and λ2 , so that Li Ai = λi Li , i = 1, 2. (6.9.3) Multiplication of (6.9.1) by Li from the left combined with (6.9.3) gives ∂ ∂ U = 0, i = 1, 2. (6.9.4) + λi Li ∂t ∂x ∂ ∂ This shows that the operators ( ∂t + λi ∂x ) represent differentiation along the respective families of characteristic curves C± defined by
C± :
dx = λi . dt
(6.9.5ab)
330
6 Kinematic Waves and Real-World Nonlinear Problems
We represent the differentiation along C+ by (6.9.4) into the form
d dξ
and along C− by
dU = 0 along C+ , dξ dU = 0 along C− . L2 dη
L1
d dη
to transform
(6.9.6) (6.9.7)
We write L1 = (L11 , L12 ) and L2 = (L21 , L22 ) to reduce (6.9.6) and (6.9.7) to the form du1 du2 + L12 = 0 along C+ , dξ dξ du1 du2 L21 + L22 = 0 along C− . dη dη L11
(6.9.8) (6.9.9)
We note that the eigenvectors L1 and L2 depend on u1 and u2 through the matrix elements aij = aij (u1 , u2 ) of A. If equations (6.9.8) and (6.9.9) are not exact, they can always be made exact equations by multiplying by the integrating factor μ. We next multiply by μ and integrate them to obtain μL11 du1 + μL12 du2 = r(η) along C+ , (6.9.10) (6.9.11) μL21 du1 + μL22 du2 = s(ξ) along C− , where r and s are arbitrary functions. These functions r(η) and s(ξ) are called Riemann invariants of the hyperbolic system (6.9.1). However, it is not necessarily the same constant on each C+ . In other words, r, in general, will vary on C− . Similarly, s is constant on each C− , but will usually vary on each C+ . In fact, the two families of characteristics are given by the equations C+ :
dx = λ1 ; dt
C− :
dx = λ2 . dt
(6.9.12ab)
Furthermore, r(η) = const. and s(ξ) = const. are shown in Figure 6.21 in which the initial data for u1 and u2 are given as follows: ˜1 (x0 ), u1 (x0 , 0) = u u1 (x1 , 0) = u ˜1 (x1 ),
u2 (x0 , 0) = u ˜2 (x0 ),
(6.9.13)
u2 (x1 , 0) = u ˜2 (x1 ).
(6.9.14)
It can be shown that, in general, a solution can be found everywhere in the (x, t)-plane at which the solution remains smooth. However, the families of characteristics C± themselves depend on the solution; hence, it is necessary to determine the solution and the characteristics simultaneously. An interesting special case exists when one of the Riemann invariants s(ξ) is identically constant, that is, s(ξ) ≡ s0 . In this case, equations (6.9.10) and (6.9.11) can be integrated to obtain
6.9 The Nonlinear Hyperbolic System and Riemann’s Invariants
331
Fig. 6.21 Characteristics C± .
f1+ (u1 ) + f2+ (u2 ) = r(η)
along C+
(6.9.15)
and
(6.9.16) f1− (u1 ) + f2− (u2 ) = s0 along C− , where fi+ (ui ) = μL1i dui and fi− (ui ) = μL2i dui . This demonstrates the fact that everywhere along a characteristic C+ specified by η = η0 = const., u1 and u2 must be constant because they are the solutions of the nonlinear system of simultaneous equations f1+ (u1 ) + f2+ (u2 ) = r(η0 ), f1− (u1 ) + f2− (u2 ) = s0 .
(6.9.17) (6.9.18)
However, solutions of this system may not be unique due to nonlinearity, and this fact is associated with possible discontinuous solutions as shock waves. This is a typical feature of nonlinear partial differential equations as explained earlier in various other problems. The solutions u1 and u2 are thus constant along the C+ characteristic specified by setting η = η0 . Hence, λ1 = λ1 (u1 , u2 ), and it follows that the characteristic must be a straight line. When one of the Riemann invariants r or s is identically constant, the corresponding solution is said to form a simple wave. Simple waves represent the simplest nonconstant solutions of a nonlinear hyperbolic system (6.9.1). The important fact that the system contains neither dispersive nor dissipative terms means that no physical mechanism exists which can prevent the formation of a discontinuous solution or that can permit the propagation of traveling waves. Example 6.9.1 Riemann’s Invariants of a Linear Wave Equation. We solve the linear initial-value problem φtt − c2 φxx = 0,
x ∈ R, t > 0,
(6.9.19)
with the Cauchy data φ(x, 0) = f (x),
φt (x, 0) = g(x)
We can factorize (6.9.19) in the form
for x ∈ R.
(6.9.20ab)
332
6 Kinematic Waves and Real-World Nonlinear Problems
∂ ∂ −c ∂t ∂x
∂ ∂ +c φ = 0. ∂t ∂x
(6.9.21)
Introducing two new functions u(x, t) and v(x, t) defined by u = φt
and v = cφx ,
(6.9.22ab)
we can combine (6.9.21) with (6.9.22ab) to obtain two first-order equations ∂ ∂ −c (u + v) = 0, (6.9.23) ∂t ∂x ∂ ∂ +c (u − v) = 0. (6.9.24) ∂t ∂x These equations imply that the quantities u±v remain constant along two families of = ±c. Further, u + v and u − v are two Riemann’s characteristics given by C± : dx dt invariants so that u + v = const. u − v = const.
dx = c, dt dx along a curve C− such that = −c. dt along a curve C+ such that
In other words, u + v = r(η) u − v = s(ξ)
along C+ : η = x − ct, along C− : ξ = x + ct,
(6.9.25) (6.9.26)
where r(η) and s(ξ) are arbitrary functions of η and ξ, respectively, to be determined from the given initial data u(x, 0) = u0 (x) and v(x, 0) = v0 (x). Along the characteristics C+ : x − ct = η and C+ : x + ct = ξ, we obtain u + v = r(η) = u0 (η) + v0 (η), u − v = s(ξ) = u0 (ξ) − v0 (ξ).
(6.9.27) (6.9.28)
Solving these equations gives the point of intersection of the lines C+ and C− on the (x, t)-plane. Clearly, both u and v are functions of ξ and η only, and hence, φ is a function of ξ and η only, by definition (6.9.22ab). Consequently, we can write φ(x, t) = Φ(ξ) + Ψ (η) = Φ(x + ct) + Ψ (x − ct),
(6.9.29)
where Φ and Ψ are arbitrary functions to be determined from the given Cauchy data (6.9.20ab). It turns out that x+ct 1 1 φ(x, t) = f (x + ct) + f (x − ct) + g(τ ) dτ, (6.9.30) 2 2c x−ct where g(x) = φt (x, 0) = u0 (x). This is the well-known d’Alembert solution of the wave equation. As indicated in Chapter 3, this method of solving the wave equation
6.9 The Nonlinear Hyperbolic System and Riemann’s Invariants
333
by using the characteristic curves C± is called the method of characteristics. This method has also been used to solve quasi-linear or nonlinear equations. The wave equation (6.9.19) with (6.9.22ab) can be expressed in the matrix form ∂U ∂U +A = 0, ∂t ∂x where
u U= v
and
A=
0 −c
(6.9.31) −c 0
.
It has been shown that the values of u and v at the point of intersection of the two characteristics C+ and C− can be found in terms of their initial values u(x, 0) = u0 (x) and v(x, 0) = v0 (x). Furthermore, the phase velocities of the wave dx dt = ±c are found to determine the families of characteristics C± . It turns out that these phase speeds are just the eigenvalues of the matrix A. To prove this, we denote the eigenvalue by λ and the corresponding right-eigenvector by R so that (6.9.32)
AR = λR.
It follows from this equation that λ = λr = ±c (r = 1, 2) and the corresponding right-eigenvectors are given by 1 1 and R− = . (6.9.33) R+ = −1 1 To investigate the eigenstates represented by R± , we write U± = φR± , where φ = φ(x, t) is an arbitrary scalar function of x and t. Substituting this in equation (6.9.31) gives the equation for φ(x, t) in the form ∂φ ∂φ +λ = 0. ∂t ∂x
(6.9.34)
Thus, the solutions of (6.9.31) corresponding to the eigenvalues λ = ±c are U+ = φ+ (x − ct)R+
and
U− = φ− (x + ct)R− .
(6.9.35ab)
They represent waves propagating to the right and left with constant velocities ±c, respectively. According to the superposition principle, the general solution of the linear wave equation is U (x, t) = U+ + U− = φ+ (x − ct)R+ + φ− (x + ct)R− .
(6.9.36)
When the initial condition U (x, 0) = U0 (x) at t = 0 is specified, equation (6.9.36) leads to (6.9.37) φ+ (x)R+ + φ− (x)R− = U0 (x). In principle, this equation can be solved, and hence, φ± (x) can be uniquely determined, since R± are linearly independent. Note that the solution represents the
334
6 Kinematic Waves and Real-World Nonlinear Problems
superposition of two suitably combined eigenstates of the matrix A involved in the equation. To obtain equations (6.9.25) and (6.9.26) involving the Riemann invariants of the system (6.9.31), we introduce the left eigenvectors L which satisfy the equation (6.9.38)
LA = λL
so that L± = (1, ∓1). Multiplying the equation (6.9.31) by L± from the left gives ∂ ∂ U± = 0. + λ± L± (6.9.39) ∂t ∂x The differential operators in (6.9.39) represent the total derivative along the characteristics C+ : x − ct = ξ and the characteristics C− : x + ct = η, corresponding to the eigenvalues λ+ = c and λ− = −c, respectively. Thus, along the characteristics ξ = x − ct = const. and η = x + ct = const., (6.9.40)
L± U± = 0.
Integrating these equations and using the eigenvectors L± = (1, ∓1), we obtain the same equations as (6.9.25) and (6.9.26). Example 6.9.2 Steady Two-Dimensional Supersonic Fluid Flow. The equations of motion for two-dimensional, irrotational, isentropic flow are 2 c − u2 ux − u v(uy + vx ) + c2 − v 2 vy = 0, (6.9.41) (6.9.42) vx − uy = 0, where u and v are fluid velocities along the x-axis and y-axis, respectively, and c is the local sound speed. In matrix notation, the preceding equations can be expressed in the form (6.9.43)
A1 Ux + B1 Uy = 0, where
A1 =
0 c2 − u2
u U = . v
1 −uv
,
B1 =
−1 −uv
0 c2 − v 2
,
and (6.9.44)
−1 2 2 Since A−1 1 exists provided c − u = 0, we multiply (6.9.43) by A1 from the left to obtain (6.9.45) Ux + A Uy = 0,
where
6.9 The Nonlinear Hyperbolic System and Riemann’s Invariants
A=
− c22uv −u2 −1
c2 −v 2 c2 −u2
335
0
.
The eigenvalues of the problem are the solutions of the equation |A−λI| = 0, which has the explicit form − 2uv c2 −v 2 c2 −u2 c2 −u2 = 0, −1 −λ that is, λ2 +
2uv c2 − v 2 λ + = 0, c2 − u2 c2 − u2
(6.9.46)
which gives two roots λ1,2 =
(c2
1 −uv ± c u2 + v 2 − c2 . 2 −u )
(6.9.47)
Clearly, the roots are real provided that q 2 = u2 + v 2 > c2 , and hence, the equations (6.9.41) and (6.9.42) are hyperbolic. The local Mach number is defined by M = q/c which is greater than unity ensuring that the flow is supersonic. In supersonic flow, the characteristics in the (x, y)-plane are defined by the equations dy = λr , dx
r = 1, 2.
(6.9.48)
These characteristics are called the Mach lines and can be determined provided the 2 2 matrix A−1 1 is nonsingular. This simply means that |A1 | = 0 or c − u = 0. If this matrix is singular, then u = c, that is, the x-component of the fluid velocity is equal to the local sound speed. The left eigenvalues corresponding to the eigenvalues λ1,2 given by (6.9.47) are L1 = (1, λ2 )
and L2 = (1, λ1 ).
(6.9.49ab)
It follows from equations (6.9.8), (6.9.9) that the characteristics in the (u, v)plane are given by ∂v ∂u + λ2 = 0; ∂ξ ∂ξ ∂v ∂u + λ1 = 0, Γ− : ∂η ∂η
Γ+ :
(6.9.50a) (6.9.50b)
where ξ and η are parameters. In terms of these parameters, equation (6.9.48) can be written as ∂x ∂y − λ1 = 0; ∂ξ ∂ξ ∂y ∂x − λ2 = 0. C− : ∂η ∂η C+ :
(6.9.51a) (6.9.51b)
336
6 Kinematic Waves and Real-World Nonlinear Problems
Combining (6.9.50a) and (6.9.51a) or (6.9.50b) and (6.9.51b) gives dy du =− , dv dx
(6.9.52)
or equivalently, dy dv · = −1. (6.9.53) dx du This is the condition of orthogonality of two curves in the (x, y)-plane and (u, v)plane when they are represented in the same coordinate plane. More precisely, C− and Γ+ characteristics and C+ and Γ− characteristics are mutually orthogonal. The system of equation (6.9.1) or (6.9.45) is called reducible provided the elements of the coefficient matrix A depend explicitly only on the elements of U . The name originates from the fact that the system can be reduced to a linear system in the new independent variables u and v by interchanging the dependent and independent variables. Such a transformation exists provided the Jacobian J = ux vy − uy vx is nonzero. Thus it turns out that ∂u ∂y =J , ∂x ∂v
∂v ∂x =J , ∂y ∂u
∂v ∂y = −J , ∂x ∂u
∂u ∂x = −J . ∂y ∂v
(6.9.54)
This transformation is called the hodograph transformation. For more details, the reader is referred to Courant and Friedrichs (1948). Example 6.9.3 Equations for the Current and Potential in a Transmission Line. We recall the equations for the current I(x, t) and potential V (x, t) from Problem 29 in 1.15 Exercise as It + Vx + RI = 0, (6.9.55) CVt + Ix + GV = 0, In matrix notation, these become (6.9.56)
A1 Ut + A2 Ux + A3 U = 0, where U=
I V
,
A1 =
L 0
0 C
,
A2 =
0 1 1 0
,
A3 =
R 0
0 G
.
We multiply the equation (6.9.56) by A−1 1 to obtain (6.9.57)
Ut + AUx + BU = 0,
where A=
0
1 L
1 C
0
and B =
R L
0
0
G C
.
The eigenvalue equation |A − λI| = 0 gives two eigenvalues λ1,2 = ±c, where 1 c = (LC)− 2 . Clearly, the velocities of the two waves are
6.9 The Nonlinear Hyperbolic System and Riemann’s Invariants
C+ :
dx = λ1 = c, dt
C− :
dx = λ2 = −c. dt
337
(6.9.58ab)
The left eigenvectors L1,2 corresponding to the eigenvalues λ1,2 are given by L1 = (cL, 1) and
L2 = (cL, −1).
(6.9.59ab)
For a lossless transmission line, R = G = 0, equation (6.9.57) becomes the classical wave equation, and the Riemann invariants become cI + V = r(η) cI − V = s(ξ)
along C+ , along C− .
(6.9.60a) (6.9.60b)
However, in general, R and G are nonzero quantities, and the equations corresponding to (6.9.8), (6.9.9), which are obtained by multiplying (6.9.57) by L1 and L2 , are G along C+ , (cL)Iξ + Vξ = − RcI + V (6.9.61a) C G along C− . (cL)Iη − Vη = − RcI − V (6.9.61b) C In general, these equations are not integrable, but they can be integrated in some speG cial cases. For the Heaviside distortionless cable, R L = C = const. = k, equations (6.9.61a), (6.9.61b) become R ∂ (cLI + V ) = − (cLI + V ) ∂ξ L ∂ R (cLI − V ) = − (cLI − V ) ∂η L
along C+ ,
(6.9.62a)
along C− .
(6.9.62b)
These equations can be integrated to obtain R cLI + V = A(η) exp − ξ along C+ , L R along C− . cLI − V = B(ξ) exp − η L Adding and subtracting leads to the results R R 1 A(η) exp − ξ + B(ξ) exp − η , I= 2cL L L 1 R R V = A(η) exp − ξ − B(ξ) exp − η . 2 L L
(6.9.63a) (6.9.63b)
(6.9.64) (6.9.65)
We can identify the equations for the characteristics as ξ = x + ct and η = x − ct, and these results can be used to reorganize the exponential factors in (6.9.64) and (6.9.65), so that the final form of the solutions becomes
338
6 Kinematic Waves and Real-World Nonlinear Problems
2R 1 exp − x f (x − ct) + g(x + ct) , 2cL L 2R 1 x f (x − ct) − g(x + ct) , V = exp − 2 L I=
(6.9.66) (6.9.67)
where f and g are arbitrary functions of their arguments. Evidently, the current and the potential represent the waveforms f + g and f − g, respectively, which propagate along the Heaviside transmission line without change of shape except for the attentuation factor exp(− 2R L x). However, if the condition R G = is not satisfied, the original shape of the waveform would suffer from distorL C tion. We close this section by adding the following comment. For the case of a homogeneous system (6.9.1) with analytic initial data, in general, a smooth solution is possible for a finite time. However, the problem for the case of inhomogeneous system (6.9.57) is more complicated. In general, a smooth solution of (6.9.57) does not exist for all time. Lax (1954a, 1954b) proved the existence of solutions for a short time and generalized the result to allow for Lipschitz continuous initial conditions which arise frequently in many physical problems, including the case of a wave propagating into a constant state.
6.10 Generalized Simple Waves and Generalized Riemann’s Invariants The preceding analysis for a two-element vector U can be extended to a hyperbolic system of the general form ∂U ∂U +A = 0, (6.10.1) ∂t ∂x where U is an n × 1 column vector with elements u1 , u2 , . . . , un , and A is an n × n matrix with elements aij = aij (u1 , u2 , . . . , un ) which depends on the elements of U , and on x and t. As a generalization of a simple wave, we take a solution U of (6.10.1) that depends on only one of the elements u1 , so that U = U (u1 ). Substituting this solution in (6.10.1) yields ∂u1 dU ∂u1 = 0. (6.10.2) I+ A ∂t ∂x du1 dU This homogeneous algebraic system for the n elements of ( du ) can have only a 1 nontrivial solution provided that the determinant equation
|A − λI| = 0
(6.10.3)
∂u1 1 holds, where λ = −( ∂u ∂t )/( ∂x ). The n solutions λr (r = 1, 2, 3, . . . , n) of (6.10.3) are simply the eigenvalues of A. We denote the right eigenvectors of A by Rr , corresponding to the eigenvalues λr . If the system (6.10.1) is totally hyperbolic, then all
6.10 Generalized Simple Waves and Generalized Riemann’s Invariants
339
the eigenvalues λr are real and distinct, and there are n linearly independent eigenvalues Rr . Hence, the family of characteristics Cr is given by ∂u1 dx ∂u1 , r = 1, 2, . . . , n. (6.10.4) = λr = − Cr : dt ∂t ∂x Or equivalently, ∂u1 ∂u1 dt + dx = 0 along each Cr , ∂t ∂x so that du1 = 0, and hence, u1 (x, t) = const.
along each Cr .
(6.10.5)
Since λr = λr (u1 ), it follows that each member of the n families of characteristics represents straight lines along which U = const. On the other hand, if (6.10.4) is written in the form ∂u1 ∂u1 + λr (u1 ) = 0, (6.10.6) ∂t ∂x it follows that u1 is the solution of the nonlinear equation (6.10.6), and this solution and hence U = U (u1 ) may or may not evolve as a discontinuous solution at a point depending on the initial data and on the form of λr (u1 ). In a linear problem in which the matrix A is independent of U , λ and R become specific functions of x and t, and there are n wave modes whose phase velocities are given by the eigenvalues λ1 , λ2 , . . . , λn . In particular, if A is a constant matrix, the linear superposition principle enables us to obtain the general solution in terms of an arbitrary set of functions φr (x − λr t) in the form U (x, t) =
n
φr (x − λr t)Rr ,
(6.10.7)
r=1
where φr can be uniquely determined from the initial condition U (x, 0) = f (x) for all x. If we now set r = k and consider the kth generalized simple wave, equation (6.10.2) must be proportional to the corresponding kth right-eigenvector of A (k) (k) (k) with elements R1 , R2 , . . . , Rn so that du1 (k) R1
=
du2 (k) R2
= ··· =
dun (k)
.
(6.10.8)
Rn
These equations represent a set of n first-order, ordinary differential equations which determines the nature of the solution U across what is called the generalized λk simple wave. Integrating (6.10.8) gives (n − 1) linearly independent relations between the n elements of U . These (n − 1) invariant relations along members of the kth family of characteristics Ck are called the generalized λ(k) -Riemann invariants and are denoted by J(k) (U ) = const.
for r = 1, 2, . . . , (n − 1).
(6.10.9)
340
6 Kinematic Waves and Real-World Nonlinear Problems
These relations hold throughout the generalized λk -simple wave, and they can be used to determine the nature of the solution provided it remains differentiable. Each generalized λ(k) -Riemann invariant defines a manifold in the (u1 , u2 , (i) . . . , un )-space on the mth of which λm (k) must satisfy the condition d J(k) = 0, which can be expressed as (i)
∂J(k) ∂u1
(i)
du1 +
∂J(k) ∂u2
(i)
du2 + · · · +
∂J(k) ∂un
dun = 0.
(6.10.10)
We consider the generalized λk -simple waves with a parameter ξ to rewrite (6.10.8) as du2 dun du1 = (k) = · · · = (k) = dξ. (6.10.11) (k) R1 R2 Rn These lead to the results dui (k) = Ri dξ
for i = 1, 2, . . . , n.
Substituting these results into (6.10.9) gives an equivalent condition (k) i ∇u J(k) R = 0,
(6.10.12)
(6.10.13)
where i = 1, 2, . . . , (n − 1) and ∇u represents the gradient operator with respect to u1 , u2 , . . . , un . It is noted here that Lax (1957, 1973) utilized condition (6.10.13), i and R(k) , first to introduce the which requires the mutual orthogonality of ∇u J(k) idea of generalized λ(k) -simple waves. In the rest of this section, we follow Jeffrey (1976) to discuss briefly the quasilinear hyperbolic system of equations with spatial derivative in the form of a onedimensional divergence ∂F (U ) ∂G(U ) + = H(U ), ∂t ∂x
(6.10.14)
where U is an n × 1 column vector with elements u1 , u2 , . . . , un , and F , G, and H are n × 1 column vectors whose elements are functions of u1 , u2 , . . . , un . If the rth element of G(U ) is gr (u1 , u2 , . . . , un ), the system of equations (6.10.14) is hyperbolic if the eigenvalues of the matrix ⎡ ∂g1 ∂g1 ∂g1 ⎤ . . . ∂u ∂u1 ∂u2 n ⎢ ∂g2 ∂g2 ∂g2 ⎥ . . . ⎢ ∂u1 ∂u2 ∂un ⎥ ⎥ (6.10.15) A=⎢ ⎢ .. .. .. ⎥ . . ⎣ . . . . ⎦ ∂gn ∂gn ∂gn . . . ∂u ∂u1 ∂u2 n are real, and it has a complete set of linearly independent eigenvectors. According to the general theory of Riemann simple waves and the development of shocks, it can be shown that the system (6.10.14) has a discontinuous solution provided that the generalized Rankine–Hugoniot shock condition
6.10 Generalized Simple Waves and Generalized Riemann’s Invariants
˜ ] = [G], λ[F
341
(6.10.16)
˜ as is satisfied, where [X] denotes the discontinuous jump in the quantity X with λ the velocity of propagation. Clearly, (6.10.16) is a nonlinear system of n algebraic ˜ the n elements of U− on the one side of the jump discontinuequations relating λ, ity, and the n elements U+ on the other side. Due to nonlinearity, a solution of this system may not be unique. In order to establish a unique solution of physical interest from the set of mathematically possible solutions, Lax (1957) generalized the selection principle for shock waves, which states that if the kth characteristic field is genuinely nonlinear, a discontinuous jump solution is admissible only if the Lax entropy criterion ˜ > λ(k) (U+ ) λ(k) (U− ) > λ (6.10.17) is satisfied. Physically, this is equivalent to the requirement that entropy cannot decrease. In fluid dynamics, this corresponds to the fact that a shock is associated with a flow which is supersonic at the front of the shock and subsonic at the rear. This leads to the physical fact that a hydrodynamical shock is a compressive wave, and it also implies an irreversible condition of the system. It has already been indicated that, in gas dynamics, this criterion is equivalent to the second law of thermodynamics which states that the entropy cannot decrease. Another important fact is that, only ˜ is equal to an eigenvalue of the matrix A, and then the jump in very special cases, λ discontinuous solution is referred to as a characteristic shock. In conclusion, we recall the generalization of a classical solution (see Chapter 5, Section 5.4) to a weak solution to resolve the difficulty involved in the breakdown of differentiability in quasi-linear hyperbolic systems. In the simplest physical problem described by the nonlinear scalar equation ∂ ∂u + f (u) = 0, ∂t ∂x the function u = u(x, t) is called the weak solution if, for t > 0, ∞ ∞ uφt + f (u)φx dx dt = 0 0
(6.10.18)
(6.10.19)
−∞
for every test function φ(x, t) ∈ C 2 which vanishes outside some finite region of the half plane t > 0. It has already been shown in Chapter 5, Section 5.4 that both discontinuous classical and weak solutions satisfy the generalized Rankine–Hugoniot condition, and a piecewise C 1 weak solution is also a piecewise C 1 classical solution. However, a weak solution is not uniquely determined by the initial data. So, the Lax entropy criterion is used to ascertain a unique and physically realistic solution. It is important to point out that weak solutions suffer from a major weakness that a variable change in the original conservation equation leads to different weak solutions. For more information about weak solutions and shock waves, we refer to Smoller (1994) and Le Veque (1990).
342
6 Kinematic Waves and Real-World Nonlinear Problems
6.11 The Lorenz System of Nonlinear Differential Equations and Deterministic Chaos In an effort to make an accurate prediction of complicated weather forecasting on Earth, Edward N. Lorenz (1917–) published a seminal paper (1963a, 1963b, 1963c) and discovered a system of three nonlinear ordinary differential equations for the three state variables (x(t), y(t), z(t)) that can exhibit chaotic behavior. Historically, the Lorenz system is one of the earliest examples of chaos realized on an electronic computer. The following are the dynamical equations of the Lorenz system dx = σ(y − x), dt
dy = rx − (y + xz), dt
dz = xy − bz, dt
(6.11.1)
where σ, r, and b are parameters. The right-hand sides of these equations do not include time t, and the time derivatives on the left-hand sides are determined solely by the state (x, y, z). Integrating the above system (6.11.1) numerically, the trajectory (x(t), y(t), z(t)) of state can be determined. A set of point in the phase space (x, y, z) where a family of trajectories for a set of initial conditions accumulates asymptotically as t → ∞ is known as the Lorenz attractor. Obviously, x, ˙ y, ˙ z˙ can be considered as three components of a velocity vector u defined at the point x = (x, y, z) so that we can regard this as a dynamical system like a fluid motion with velocity field u(x). One of the important features of such a motion in the phase space is that the phase volume consisting of points moving with velocity u = (x, ˙ y, ˙ z) ˙ decreases steadily because the divergence of the velocity field is negative, that is, div u =
∂ y˙ ∂ z˙ ∂ x˙ + + = −(σ + 1 + b) < 0. ∂x ∂y ∂z
(6.11.2)
This means that the phase volume of an attractor where the trajectories are approaching decreases indefinitely. This does not necessarily imply that the attractor is made of discrete points, but means only that the dimension of the attractor is less than three. Thus, the dimension of the attractor becomes non-integral or fractal. Lorenz made a serious attempt to solve an initial value problem of the system (6.11.1) with σ = 10, r = 28, b = 83 , and the initial data (x(0), y(0), z(0)) = (0, 1, 0). His numerical experiments show that the Lorenz system (6.11.1) captured a certain new but intrinsic chaotic nature of the weather phenomena, and the long-term weather prediction is almost impossible. It also revealed that a very small change in initial conditions leads to a significantly large change in the solution. It is now a well known fact that the temporal behavior of the Lorenz system is stochastic in a true sense. In fact, its short time behavior is deterministic, but its long time evolution is stochastic. This phenomenon is known as the deterministic chaos. Indeed, many nonlinear dynamical systems have property similar to that of the Lorenz attractor. Included is Figure 6.22 for a Lorenz attractor which is a representative example of the strange attractor in nonlinear dynamics. In stability analysis, steady state plays a fundamental role because the decay or growth of perturbations of a steady state is usually investigated for a dynamical
6.11 The Lorenz System of Nonlinear Differential Equations
343
Fig. 6.22 A numerically computed solution of the Lorenz equations with a plot of z against x with r = 28, (x(0), y(0), z(0) = (5, 5, 5)).
system. A steady state corresponds to a fixed point of a dynamical system. In case of the Lorenz system (6.11.1), a fixed point is obtained from (x, ˙ y, ˙ z) ˙ = (0, 0, 0). Thus, it follows from system of equations (6.11.1) that x = y, rx − (y + xz) = 0, xy − bz = 0. For r < 1, there is only one fixed point which is the origin O : (x, y, z) = (0, 0, 0). The origin O corresponds to the static state. This is in full agreement with the analysis of thermal convection problem in fluid dynamics where we consider a bifurcation problem from the static state to a thermal convection-cell state. However, for r > 1, it can be shown that there exist three fixed points: the originO and two points given by C1 (q, q, r − 1) and C2 (−q, −q, r − 1) where q = b(r − 1), and C1 and C2 are located at points of mirror symmetry with respect to the vertical plane x + y = 0. When r > rc (rc > 1), all three fixed points are unstable, that is, there are no stable fixed points to which the trajectory can approach. This is really a strange phenomenon. In order to study the stability of the origin O, we first linearize the Lorenz system for points in the neighborhood of O so that dx = σ(y − x ), dt
dy = rx − y , dt
dz = −bz , dt
(6.11.3)
where x , y , z are small perturbations to the fixed point (0, 0, 0), which are represented as (x , y , z ) = (x0 , y0 , z0 )eλt , where λ is the growth rate. Substituting these perturbations in (6.11.3) leads to a system of linear algebraic equations in the matrix form for the amplitude (x0 , y0 , z0 ): ⎛ ⎞⎛ ⎞ λ + σ −σ 0 x0 ⎝ −r λ+1 0 ⎠ ⎝ y0 ⎠ = 0. (6.11.4) z0 0 0 λ+b
344
6 Kinematic Waves and Real-World Nonlinear Problems
For a nontrivial solution of (x0 , y0 , z0 ), the determinant of the coefficient matrix must vanish. Thus, it gives the eigenvalue equation for the growth rate λ, that is, λ + σ −σ 0 −r λ+1 0 = (λ + b) λ2 + (σ + 1)λ + σ(1 − r) = 0. (6.11.5) 0 0 λ + b Consequently, the eigenvalues are given by λ = −b,
λ± =
1 −(σ + 1) ± (σ + 1)2 + 4σ(r − 1) . 2
(6.11.6)
For positive σ and b, we draw the following conclusions: For 0 < r ≤ 1, all three eigenvalues are real and non-positive. Hence, the fixed (or equilibrium) point O is stable. For r > 1, two eigenvalues are real and negative, but one eigenvalue is real and positive. Thus, O is unstable, in general. When r = 1, it is a transition (or bifurcation) point. The fixed point O serves as the transition point from stability to instability, and at the same transition point, new equilibrium points begin to appear. These results are both qualitatively and quantitatively in agreement with the linear stability analysis of the Rayleigh–Benard convection problem in fluid mechanics. It has been shown that the situation r > rH = 27.74 is quite complicated. It can be shown that there exists a stable periodic orbit when r is very large. We introduce a small parameter ε = √1r and time scale t = εt . We next substitute x x= , ε
y=
1 y , ε2 σ
1 z= 2 ε
z +1 , σ
(6.11.7)
in the Lorenz system (6.11.1) so that it becomes, dropping the primes, dx = y − εσx, dt
dy = −(xz + εy), dt
dz = xy − εb(z + σ). dt
(6.11.8)
In the limit as r → ∞, ε → 0 so that (6.11.8) reduces to dx = y, dt
dy = −xz, dt
dz = xy. dt
(6.11.9)
Consequently, dx dy dz = = = dt y −xz xy
(6.11.10)
lead to two integrals representing a periodic orbit in the form x2 − 2z = 2A,
y2 + z2 = B2 ,
(6.11.11)
where A and B are constants. Thus, the trajectories lie on the cylinder y2 + z 2 = B 2 and, in general, it is a closed orbit. It can be shown based on perturbation expansion for small ε that there exists a stable periodic orbit when (σ + 1) > (b + 1). The orbit
6.11 The Lorenz System of Nonlinear Differential Equations
345
winds once around the z-axis for large r, this is the only orbit which persists for all r. In the case of the Lorenz system for σ = 10 and b = 83 , the above condition (σ + 1) > (b + 1) is satisfied. Numerical computations reveal that as r decreases, period-doubling bifurcation will begin to occur at r = 313. Successive stages of period-doubling bifurcations will occur as r decreases and reaches the value 214.364. At each period-doubling bifurcation, the original periodic orbit becomes unstable, and new stable orbits which wind twice as much as the original will appear. For such phenomena of period-doubling, if rn represents the value of r at the nth bifurcation, then, in the limit as n → ∞, the celebrated Feigenbaum constant, F , appears. More precisely, rn−1 − rn F = lim = 4.6692016609 . . . . (6.11.12) n→∞ rn − rn+1 This is known as a universal (probably transcendental) constant which often arises in many problems in nonlinear dynamics. In the numerical experiment of the Lorenz system with σ = 10, r = 28, and b = 83 , two orbits beginning from two neighboring points near Lorenz attractor lose their correlation after some time. Such a property of sensitive dependence on initial conditions is a characteristic feature of chaotic orbits. After some time, both orbits trace a path on the same attractor without mutual correlation. The Lyapunov dimension (or fractal dimension), DL , of the Lorenz attractor for the above parameter values is found as 2.00 < DL < 2.401. In view of the fractal dimension, the Lorenz attractor is called a strange attractor. Historically, chaos is the revelation of the Lorenz (strange) attractor. Chaotic flow is irregular and appears to be random in nature, but it is deterministic. In fact, the deterministic chaos arises in many different nonlinear physical systems. Chaos may represent a state leading to turbulence. From a modern view point, turbulence is regarded as a highly irregular fluctuating flow field which develops autonomously by a nonlinear mechanism of field dynamics. This change in view of turbulence is mainly due to the recent development of the theory of chaos. From 1960s, numerical experiments have confirmed the existence of phenomena such as chaos, intermittency and period-doubling process. As far as the Lorenz nonlinear system of equations is concerned, it represents at best a mathematical model for the Rayleigh–Benard convection flow problem when the Reynolds number R is only sufficiently large compared to the critical value Rc . On the other hand, it follows from the numerical studies of the Lorenz system of equations that a geometric model can be constructed, which can be defined by a nonlinear map on an interval. If we follow a trajectory given by the Lorenz equations and consider the plane z = r − 1, the trajectory will cross from the side z > r − 1 to the other side at various points on z = r − 1. The successive points on the plane section z = r − 1 define a return map which is known as the Poincaré return map. This shows that there is an interesting relationship between the Lorenz system of differential equations and the nonlinear map on an interval as can be illustrated by the famous logistic map. The logistic map is defined by the following recurrence relation for an integer n:
346
6 Kinematic Waves and Real-World Nonlinear Problems
xn+1 = λf (xn ) = λxn (1 − xn ),
(6.11.13)
where xn ∈ [0, 1] and λ ∈ (0, 4]. It is to be noted that the logistic map illustrates general features of nonlinear maps. This map is one of the simple models for the growth of the population of a single species. With an initial population x0 , equation (6.11.13) represents a measure of the population in subsequent generations. The state xn = 0 gives the complete absence of the species, and when xn 1, xn+1 ∼ λ xn so that the generation grows by a factor of λ. If xn is not small, the term (1−xn ) is not close to one and the full nonlinear equation (6.11.13) determines the size of the next generation. If x0 ∈ [0, 1], then 0 ≤ xn ≤ 1 for n ≥ 0. Thus, the interval [0, 1] is the physically meaningful range of the map. The fixed points of (6.11.13) satisfy xn+1 = xn so that xn = λxn (1 − xn ). So, it has two fixed points at x = 0 and x = ( λ−1 ). λ We next determine whether there are any solutions of period 2 (or 2-cycles) so that they satisfy xn+1 = λxn (1 − xn ) and xn+2 = xn = λxn+1 (1 − xn+1 ). Thus, the elimination of xn+1 leads to the equation for xn , λ−1 2 2 λ xn − λ(1 + λ)xn + (1 + λ) = 0. xn xn − (6.11.14) λ This equation must have two fixed points. The discriminant of the quadratic factor in (6.11.14) is λ2 (λ + 1)(λ − 3) which is positive for λ > 3. So, there are fixed points of period 2 (or 2-cycle) for λ > 3. In fact, it can be shown that the nontrivial fixed point is stable for λ < 3, but loses stability in a bifurcation at λ = 3, where the points of period 2 emerge and are stable. As λ increases, the points of period 2 eventually lose stability, and a stable of 4-cycle emerges. This process is known as period-doubling, and as λ increases, it eventually leads to chaotic solutions. This phenomenon of period-doubling has become famous in recent years. It is possible to find the successive values λ1 , λ2 , λ3 , . . . of the growth rate parameter at which a bifurcation or qualitative change in the iteration xn+1 = λxn (1 − xn ) occurs as the value of λ increased further. These are the discrete values of λ at which any sufficiently small increase in λ doubles the period of the iteration. This also leads to a certain order underlying this period-doubling process toward chaos with the same Feigenbaum constant F in (6.11.12). This map takes its name from the corresponding differential equation dx = μx(1 − x), μ > 0. (6.11.15) dt This equation was originally suggested by Pierre-Francois Verhulst (1804–1849) in the study of population dynamics. It is worth noting that there are two equilibrium (or fixed) points of (6.11.15), one at x = 0 which is unstable, and another at x = 1 which is stable. However, the behavior of (6.11.13) which is the difference equation version of differential equation (6.11.15) is quite different.
6.12 Exercises 1. If h(x, t) is the depth of the water level above impermeable rock and if the horizontal seepage velocity u and the slope of the water level satisfy the Darcy
6.12 Exercises
347
law, u = −κ hx , where κ is a constant, show that the mass conservation equation is given by ht = κh2x + κhhxx . 2. In Example 6.3.2, the discontinuous initial data can be approximated by continuous data in the form ⎧ if x < −ε, ⎪ ⎨ 200 x ρ(x, 0) = f (x) = 100(1 − ε ) if − ε < x < ε, ⎪ ⎩ 0 if x > ε. Find the solution for ρ(x, t) = f (ξ), with the value of x in terms of ξ and t. Express the solution ρ in the terms of x and t, and draw the graph of ρ(x, 0). 3 (200 − ρ)ρ, 3. (a) Solve the traffic flow equation (6.3.4) with the flow q(ρ) = 10 and with data representing a triangular hump whose center is at the origin given by 25(3 − |x|), if |x| ≤ 1, ρ(x, 0) = f (x) = 50 if |x| > 1. (b) Examine the time evolution of the density profile at three different times 1 1 1 (1 − ε), 15 , and 15 (1 + ε), for 0 < ε < 1. Show that the solution breaks t = 15 1 1 ). down for t = 15 . Draw the graph of the solution ρ(x, 15 4. Show that any solution of the traffic flow equation ρt + c0 ρx = aρ, where c0 and a are constants, has the form ρ(x, t) = exp(at)f (x − c0 t). 5. Find the solution of the traffic flow model ρt + c(ρ)ρx = 0, where c(ρ) = 60 − 35 ρ, with initial data 150{1 + 15 (1 − |x|)} ρ(x, 0) = 150
if |x| ≤ 1, if |x| > 1
6. Obtain the solution of the traffic flow equation ρt + c(ρ)ρx = 0,
ρ(x, 0) = 50H(−x),
where c(ρ) = 60 + 35 ρ. 7. For a traffic flow model with flow q = Aρ(ρm − ρ) − κρx , where A, ρm , and κ are constants, show that the traffic density satisfies the equation ρt + A ρ(ρm − ρ) x = κρxx . Using the nondimensional variables
348
6 Kinematic Waves and Real-World Nonlinear Problems ∗
x =
−1
x,
∗
t =
Aρm
−1
t,
and u =
1 ρm 2
−1
1 ρm − ρ , 2
where is a characteristic length scale associated with the initial data, show that the Burgers equation, dropping the asterisks, is given by ut + uux = vuxx , where v = ( A κρm )/ is the dimensionless parameter which is the ratio of the diffusion length and the characteristic length . 8. For a traffic flow model with the quadratic flow q(ρ)in the region 0 < ρ < ρm and zero otherwise, that is, ρ ρ q 1− , 0 < ρ < ρm , = qm ρm ρm where ρm is the maximum traffic density, show that ρ(x, t) satisfies the equation ρt + c(ρ)ρx = 0, ) and um = ρqm . where c(ρ) = um (1 − ρ2ρ m m 9. The concentration C(x, t) of a cloud of uniform sedimenting particles in a viscous fluid satisfies the equation (Kynch 1952) ∂ ∂C + v0 C(1 − αC) = 0, ∂t ∂x where v0 and α are constants. If the bottom at x = 0 is impermeable and the initial concentration is C0 for all x < 0, where C0 is a constant, discuss the changes of C(x, t) for all x < 0 and t > 0. 10. Apply the method of characteristics to solve the system of equations ∂u ∂v + = 0, ∂t ∂x
∂v ∂u + =0 ∂t ∂x
with the initial data u(x, 0) = ex
and v(x, 0) = e−x .
Show that the Riemann invariants are 2r(ξ) = 2 cosh ξ and 2s(η) = 2 sinh η. Hence, show that the solutions are (u, v) = cosh(x − t) ± sinh(x + t). 11. The shallow water equations are ut + uux + ghx = 0, ht + uhx + hux = 0,
6.12 Exercises
349
where u = u(x, t) is the fluid velocity and h = h(x, t) is the height of the free surface above the horizontal bottom. If u = u(h), show that ∂u g =± , u = ±2(c − c0 ), ∂h h √ √ where c = gh, c0 = gh0 , and h0 is the equilibrium value of h. Find the Riemann invariants of this system. 12. For one-dimensional anisentropic flow, the Euler equations of motion are pt + upx + c2 ρux = 0, 1 ut + uux + px = 0, ρ St + uSx = 0, where p = f (ρ, S), u is the velocity, ρ is the density, p is the pressure, and S is the entropy. Show that this system has three families of characteristics given by C0 :
dx = u, dt
C+ :
dx = u + c, dt
and C− :
dx = u − c, dt
where c2 = ( ∂p ∂ρ )x = const. 13. Show that the second-order equation aφtt + 2bφxt + cφxx = 0, where a, b, c are real functions of φx and φt , can be reduced to the system in the 2b c a a . matrix form (6.9.31) where A = −1 0 Show that eigenvalues λ are the roots of the quadratic equation a λ2 − 2bλ + c = 0. Hence, show that the left eigenvectors are L1,2 = (1, λ2,1 ), where the indices 1 and 2 in L correspond to the indices 2 and 1 in λ, respectively. 14. If a, b, and c are constants and a2 = 4b, then the linear telegraph equation utt − c2 uxx + aut + bu = 0 can be factorized into two first-order equations. Discuss the characteristics of the two first-order equations. 15. An observer at a position x = x(t) moves with the traffic. Show that the rate of the change of the traffic density measured by the observer is given by dq ∂ρ dρ = u− , dt dρ ∂x where u is the velocity of the car.
350
6 Kinematic Waves and Real-World Nonlinear Problems
16. Show that equations (6.8.106) and (6.8.107) can be written as Ut + A Ux = B, where
u U= , c
A=
u
2c u
c 2
,
and B =
Hx 0
.
Show that the eigenvalues are λ1,2 = u ± c and the corresponding left eigenvectors are L1 = (1, 2) and L2 = (1, −2). Derive the equations for the directional derivatives of (u ± 2c) along the C± characteristics. Hence, find these equations when Hx = gh = const. = m. 17. Show that the nonlinear wave equation c2 (φx )φxx − φtt = 0, where c is an even function of φx , can be written as Ut + A Ux = 0, where
u φx , U= = φt v
A=
0 −c2
−1 0
,
and c = c(u).
Hence, show that the eigenvalues λ1,2 and the corresponding left eigenvectors L1,2 are given by λ1,2 = ∓c
and L1,2 = (c, ±1).
Show that the Riemann invariants are r(η) and s(ξ) given by m(u) + v = r(η) along dx = −c, and m(u) − v = s(ξ) along dx = +c, where m(u) is defined dt dt by m(u) = c(u) du. 18. Express (Jeffrey and Taniuti 1964) equations (6.8.16) and (6.8.17) in the matrix form Ut + AUx = 0, where U and A are given by (6.8.19). Show that the eigenvalues λ1,2 and the corresponding left eigenvectors are given by λ1,2 = u ± c and L1,2 = ( ρc , ±1). Hence, show that u + m(ρ) = r(η)
dx = u + c, dt dx = u − c, along C− : dt
along C+ :
u − m(ρ) = −s(ξ) where m(ρ) is defined by
m(ρ) =
c(ρ) dρ. ρ
Discuss the simple wave solution and discontinuity at the cusp of the envelope of characteristics.
6.12 Exercises
351
19. Using Example 6.8.3 in a fluid of constant depth (Hx = 0), show that the Riemann invariants r and s are u + 2c = r(η) u − 2c = s(ξ)
dx = u + c, dt dx along characteristics C− : = u − c. dt along characteristics C+ :
Hence, derive 1 1 (r + s), c = (r − s), 2 4 dt dx dt = (u + c) , = (u − c) , dξ dη dη 1 ∂t 1 ∂t ∂x = (3r + s) and C− : = (r + 3s) . 4 ∂s ∂r 4 ∂r
u= dx dξ ∂x C+ : ∂s
Use the last two results to obtain the Euler–Poisson–Darboux equation for t(r, s) in the form 2(r − s)trs + 3(ts − tr ) = 0. 20. Use Exercise 16 to show that equations (6.8.106) and (6.8.107) can be written in the conservation form 1 2 ∂ ∂ u u + c2 − H(x) 2 + = 0. u c2 ∂t c2 ∂x 21. Show that the Tricomi equation uxx + xuyy = 0 has the matrix form Ux + AUy = 0,
where u=
v w
and
A=
0 −1
x 0
.
Solve the eigenvalue equation |A − λI| = 0 to show that the Tricomi equation is hyperbolic, parabolic, or elliptic accordingly as x < 0, x = 0, or x > 0. 22. The equations of motion for the unsteady isentropic compressible fluid flows are ∂ρ + div(ρ u) = 0 ∂t and
1 ∂u + (u · ∇)u + ∇p = 0, ∂t ρ
dp where u = (u, v, w) is the flow field, p = p(ρ), and c2 = dρ is the square of the local sound speed. Express these equations in the matrix form
352
6 Kinematic Waves and Real-World Nonlinear Problems
Ut + AUx + BUy + CUz = 0, where ⎛
⎞ ρ ⎜u⎟ ⎟ U =⎜ ⎝ v ⎠, w
⎛
u ⎜ c2 /ρ A=⎜ ⎝ 0 0
⎞ 0 0⎟ ⎟, 0⎠ u
ρ 0 u 0 0 u 0 0
⎛
and
w 0 ⎜ 0 w C=⎜ ⎝ 0 0 c2 /ρ 0
⎛
v 0 ⎜ 0 v B=⎜ ⎝ c2 /ρ 0 0 0
ρ 0 v 0
⎞ 0 0⎟ ⎟, 0⎠ v
⎞ ρ 0⎟ ⎟. 0⎠ w
0 0 w 0
Show that the equation of motion has the conservation form ∂F + div G = 0, ∂t where
⎛
⎞ ρ ⎜ ρu ⎟ ⎟ F =⎜ ⎝ ρv ⎠ ρw
⎛
and
ρu ⎜ p + ρu2 G=⎜ ⎝ ρuv ρuw
ρv ρuv p + ρv 2 ρvw
⎞ ρw ρuw ⎟ ⎟. ρvw ⎠ p + ρw2
23. In the one-dimensional case of the system of equations in Exercise 22, show that the matrix form of the system is Ut + A Ux = 0, where
ρ U= u
and A =
u ρ c2 /ρ u
.
24. Reduce the second-order nonlinear wave equation for the displacement y(x, t) of a string −2 2 c = T ∗ /ρ ytt = c2 1 + yx2 yxx to the first-order system of equations −2 vx , ut = c2 1 + v 2
vt = ux ,
where u = yt and v = yx . Express this system in the matrix form Ut + AUx = 0, where
u U= v
and A =
0 −1
−c2 (1 + v 2 )−2 0
.
6.12 Exercises
353
25. Show that the nonuniqueness of the method of reduction can be seen for the equation autt + cuxx + u = 0 to the system Ut + AUx + B = 0, when the two following equivalent substitutions are used: (i) v = ut ,
w = ux
and
(ii) v = ut ,
w = ux + ux .
Show also that the equation |A − λI| = 0 gives in both cases the genuine eigenvalues λ = ± −c/a, for (i) the redundant eigenvalue λ = 1, and for (ii) the redundant eigenvalue λ = 12 . 26. Traffic velocity is related to the traffic density by u = u(ρ) = um exp(−ρ/ρ0 ), where um is the maximum traffic velocity and ρ0 is a reference traffic density. Solve the initial-value problem ρt + c(ρ)ρx = 0, x ∈ R, t > 0, ⎧ if x < 0, ⎪ ⎨ ρ0 ρ(x, 0) = ρ0 (a − x) if 0 < x < a, ⎪ ⎩ 0 if x > a, d where c(ρ) = q (ρ) = dρ [ρ u(ρ)]. Plot x against (ρ/ρ0 ) for 0 < ρρ0 < a and a fixed t. 27. For the one-dimensional case of the system of equations in Exercise 22, show that (a) the scalar conservation form of the system is (ρu)t + ρu2 + p x = 0,
(b) the matrix conservation form is Ut + F (U ) x = 0,
where U=
ρ ρu
and
F (U ) =
ρu . ρu2 + p
(c) If ρ = ρ(u), the equations in Exercise 22 in 1D can be written in the form dρ ∂u ∂ρ ∂u + ρ+u = 0, du ∂t ∂v ∂x c2 (ρ) dρ ∂u ∂u + u+ = 0. ∂t ρ du ∂x
354
6 Kinematic Waves and Real-World Nonlinear Problems
28. Use ut = v, ux = w to transform the Tricomi equation xutt + uxx = 0, in the matrix form Ut + AUx = 0, where v 0 U= and A = −1 w
1 x
0
.
29. The equations for long wave approximation for nonlinear shallow water waves are ht + uhx + hux = 0 and ut + uux + ghx = 0, where h is the depth of water, and u is the horizontal fluid velocity and g is the acceleration due to gravity. Express these equations in the conservation form 1 2 2 (uh)t + hu + gh = 0. 2 x
7 Nonlinear Dispersive Waves and Whitham’s Equations
The tool which serves as intermediary between theory and practice, between thought and observation, is mathematics; it is mathematics which builds the linking bridges and gives the ever more reliable forms. From this it has come about that our entire contemporary culture, in as much as it is based on the intellectual penetration and the exploitation of nature, has its foundations in mathematics. Already Galileo said: one can understand nature only when one has learned the language and the signs in which it speaks to us; but this language is mathematics and these signs are mathematical figures. . . Without mathematics, the astronomy and physics of today would be impossible; these sciences, in their theoretical branches, virtually dissolve into mathematics. David Hilbert
7.1 Introduction Historically, the study of nonlinear dispersive waves started with the pioneering work of Stokes in (1847) on water waves. Stokes first proved the existence of periodic wavetrains which are possible in nonlinear dispersive wave systems. He also determined that the dispersion relation on the amplitude produces significant qualitative changes in the behavior of nonlinear waves. It also introduces many new phenomena in the theory of dispersive waves, not merely the correction of linear results. These fundamental ideas and the results of Stokes have provided a tremendous impact on the subject of nonlinear water waves, in particular, and on nonlinear dispersive wave phenomena, in general. Stokes’ profound investigations on water waves can be considered as the starting point for the modern theory of nonlinear dispersive waves. In fact, most of the fundamental concepts and results on nonlinear dispersive waves originated in the investigation of water waves. The study of nonlinear dispersive waves has proceeded at a very rapid pace with remarkable developments over the past three decades. L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_7,
356
7 Nonlinear Dispersive Waves and Whitham’s Equations
This chapter is devoted to a general treatment of linear and nonlinear dispersive waves. The initial-value problems of linear dispersive waves and their asymptotic analysis are briefly described. Included are Whitham’s theory of nonlinear dispersive waves, Whitham’s averaged variational principle, Whitham’s equation, the peaking and breaking of waves, and Whitham’s nonlinear instability analysis and its application to water waves.
7.2 Linear Dispersive Waves We consider a dynamical problem so that a small disturbance φ(x, t) with reference to the undisturbed stable state is governed by a linear partial differential equation with constant coefficients: ∂ ∂ ∂ ∂ φ(x, t) = F (x, t), (7.2.1) , P , , ∂t ∂x1 ∂x2 ∂x3 where P is a polynomial, and t and x = (x1 , x2 , x3 ) = (x, y, z) are independent time and space variables. The term F (x, t) represents the action of external forces on the dynamical system and is usually referred to as a given forcing term. We seek a plane wave solution of the homogeneous equation ∂ ∂ φ(x, t) = 0 (7.2.2) , P ∂t ∂xi in the form
κ · x − ωt) , φ(x, t) = A exp i(κ
(7.2.3)
where A is the amplitude, κ = (k1 , k2 , k3 ) ≡ (k, l, m) is the wavenumber vector, and ω is the frequency. Such a solution exists provided that the dispersion relation P (−iω, ik1 , ik2 , ik3 ) ≡ D(ω, κ ) = 0
(7.2.4)
is satisfied. Thus, there exists a direct relationship between the governing equation and the dispersion relation through the obvious correspondence ∂ ↔ −iω, ∂t
∂ ↔ ikj . ∂xj
(7.2.5ab)
This correspondence allows us to obtain the dispersion relation from the governing equation (7.2.2), and conversely, equation (7.2.2) can be constructed from the given dispersion relation (7.2.4). If, in particular, the dispersion relation (7.2.4) can be solved explicitly in terms of real roots given by κ) = Ω(ki ), (7.2.6) ω = Ω(κ then there may be several possible roots for the frequency, corresponding to different modes of propagation.
7.2 Linear Dispersive Waves
357
The phase velocity of the waves is the velocity at which a surface of constant phase moves. It is defined by the relation c=
ω , κ κ
(7.2.7)
is the unit vector in the direction of the wavenumber vector κ . where κ κ), the phase velocity depends on the Thus, for any particular wave mode ω = Ω(κ wavenumbers k1 , k2 , and k3 . In other words, different waves propagate with different phase velocities, and such waves are called dispersive provided the determinant |∂ 2 Ω/∂ki ∂kj | = 0. On the other hand, waves are called nondispersive if the phase velocity does not depend on the wavenumber. For one-dimensional dispersive waves, the determinant reduces to Ω (k) = 0. However, in general, the governing dispersion relation for one-dimensional waves is ω = Ω(k), which may give a complex ω for a real k. In such a case, the phase velocity depends not only on wavenumber, but also on the amplitude of the waves. This means that the amplitude will grow or decay in time accordingly as Im (ω) > 0 or < 0. So, the former case leads to instability. The group velocity vector for the three-dimensional wave motion is defined by the result ∂ω ∂ω ∂ω κ ) = ∇κ ω = . (7.2.8) , , C(κ ∂k ∂l ∂m For one-dimensional waves, the group velocity C(k) = (dω/dk). It is noted that the group velocity plays a fundamental role in the theory of dispersive waves and is much more important than the phase velocity. The following are examples of linear dispersive waves. 1. The governing dispersion relation for surface waves on water of constant depth h: ω 2 = gk tanh kh. 2. The Korteweg–de Vries (KdV) equation for long water waves: ω = c0 k − αk3 . φt + c0 φx + αφxxx = 0 c20 = gh ,
(7.2.9)
(7.2.10ab)
3. The Boussinesq equation for long water waves: φtt − c20 φxx − β 2 φxxtt = 0, −1 . ω 2 = c20 k 2 1 + β 2 k 2
(7.2.11a) (7.2.11b)
4. The Benjamin–Bona–Mahony (BBM) equation: φt − c0 φx − αφxxt = 0,
−1 ω = −c0 k 1 + αk2 .
(7.2.12ab)
1/2 ω = ± c2 κ2 + d2 .
(7.2.13ab)
5. The Klein–Gordon equation: φtt − c2 ∇2 φ + d2 φ = 0,
358
7 Nonlinear Dispersive Waves and Whitham’s Equations
6. The internal waves in a stratified ocean, governed by the equation 2 2 2 ∂ ψ ∂2ψ ∂ ∂ 2∂ ψ + = 0, + N +U ∂t ∂x ∂x2 ∂z 2 ∂x2
(7.2.14ab)
Nk ω = Uk ± √ , 2 k + m2 where ψ(x, z, t) represents the stream function and
1/2 g ∂ ρ¯ N= − ρ0 ∂z is a constant Brunt–Väisälä frequency. 7. The inertial waves in a rotating liquid, governed by the equation 2 ∂ fm ∂ ∂ 2χ ∇2 χ + f 2 2 = 0 (f = 2Ω), +U . ω = iU m ± ∂t ∂z ∂z κ (7.2.15ab) 8. The internal-inertial waves in a rotating stratified ocean, governed by the equation 2 2 ∂ ∂ f 2 m2 + N 2 (k2 + 2 ) 2 2 2 ∇ χzz = 0, + N χ + + f ω2 = , h 2 2 ∂t ∂t κ2 (7.2.16ab) where ∇2h = ∂ 2 /∂x2 + ∂ 2 /∂y2 . 9. The Rossby waves in a β-plane ocean, governed by the equation 2 f2 βk a2 = 2 , . (7.2.17ab) ∇h −a2 ψt +βψx = 0 ω=− 2 c0 k + 2 + a2 10. The dispersion relation for the Alfvén gravity waves in an electrically conducting liquid of constant depth h: ω 2 = gk + a2 k2 tanh kh, (7.2.18) √ where a = B0 / 4πρ is the Alfvén wave velocity. 11. The dispersion relation for electromagnetic waves in plasmas: ω 2 = ωp2 + c2 k 2 ,
(7.2.19)
where ωp = (4πne2 /m)1/2 is the plasma frequency, c is the velocity of light, and n is the number of electrons (per unit volume) of mass m and charge −e. 12. The Schrödinger equation and de Broglie waves: iψt = V ψ −
2 2 ∇ ψ, 2m
ω=
V κ2 + , 2m
where h(= 2π) is the Planck constant and V is a constant potential.
(7.2.20ab)
7.3 Initial-Value Problems and Asymptotic Solutions
359
The phase and the group velocities associated with these examples can readily be calculated. For dispersive waves, the former are different from the latter. The general solution for any dispersive wave system can then be obtained by the Fourier superposition of the plane wave solution (7.2.3) for different wavenumbers (k, l, m) with the corresponding frequencies ω to satisfy the dispersion relation (7.2.6). The solution is represented by ∞ ∞ ∞
κ) exp i κ · x − Ω(κ κ)t dκ κ, F (κ (7.2.21) φ(x, t) = −∞
−∞
−∞
κ) is determined from appropriate initial or boundary where the spectrum function F (κ conditions. Although (7.2.21) is the exact solution, the principal features of the dispersive waves cannot be described without an exact or approximate evaluation of the integral in (7.2.21). We shall discuss this point in the next section, which is concerned with initial-value problems. It is relevant to mention here that, in general, the governing equations for dispersive waves are inherently nonlinear, and the corresponding dispersion relation is also nonlinear in the sense that the frequency ω is not only a function of κ , but also of other parameters, such as amplitude and local properties of the medium. The theory of water waves provides an excellent example of nonlinear dispersion. In the study of linear dispersive systems, the preceding parameters are assumed to be small, and the governing equations for such systems are obtained by linearizing the original nonlinear equations. The classical theory of water waves satisfying Laplace’s equation and the linearized free surface conditions can again be cited as an excellent example of the process of linearization. However, in his pioneering work based upon averaged variational principles, Whitham (1974) gave a new description of nonlinear water waves and a completely different approach to nonlinear dispersive waves, in general. It has become clear from Whitham’s theory that nonlinear dispersion is much more important than the corresponding linearized concept and plays a significant role in the general theory of dispersive wave systems.
7.3 Initial-Value Problems and Asymptotic Solutions We consider the initial-value problem of wave propagation in which the disturbance φ(x, t) is given by f (x) at time t = 0. The general solution for any particular mode κ) is given by the threeof propagation with the known dispersion relation ω = Ω(κ dimensional integral ∞
κ) exp i κ · x − Ω(κ κ)t dκ κ, φ(x, t) = F (κ (7.3.1)
−∞
κ represents a line, area, or volume integral, depending on the number of where dκ dimensions n = 1, 2, or 3. At t = 0, result (7.3.1) gives
360
7 Nonlinear Dispersive Waves and Whitham’s Equations
∞
f (x) = φ(x, 0) = −∞
κ) exp i(κ κ · x) dκ κ, F (κ
(7.3.2)
κ) is the generalized Fourier transform of the initial disturbance f (x), so where F (κ that ∞ 1 κ) = F (κ f (x)e−iκκ·x dx. (7.3.3) (2π)n/2 −∞ κ) can be determined from the given initial data. Thus, the spectrum function F (κ Whatever the particular initial-value problem may be, the general solution can always be represented by the integral (7.3.1). The number of wave modes of type (7.3.1) depends on the particular problem. So, the complete solution will be the sum of κ). terms such as (7.3.1) with one integral for each of the solutions ω = Ω(κ Note that the physical interpretation of the integral solution (7.3.1) is not at all obvious from its present form. On the other hand, the exact evaluation of (7.3.1) is a difficult task. To investigate the principal features of dispersive waves, it is important to consider the asymptotic evaluation of (7.3.1) for large x and t. It is then necessary to resort to asymptotic methods. We next turn our attention to asymptotic evaluation of (7.3.1) by using the method of stationary phase (Jones 1966). According to this method, the main contribution to the integral (7.3.1) for large t is from the terms κ) associated with the stationary points, if any, where the phase function κ · x − tΩ(κ is stationary, that is, where κ) xi ∂Ω(κ = . (7.3.4) ∂ki t It can be shown that the dominant contribution to the integral (7.3.1) will come from those values of ki = ki (xi , t) that satisfy (7.3.4). Thus, the asymptotic solution from one such stationary point ki is obtained in the standard form 2 ∂ Ω (2π)n/2 F (ki ) π κ) − sgn , φ(x, t) ∼ exp i κ · x − tΩ(κ ∂2Ω n 1/2 4 ∂ki2 {t det | ∂ki ∂kj |} (7.3.5) where the summation convention is used. The complete asymptotic solution for φ(x, t) will be the sum of terms like (7.3.5) with one term for each stationary point ki of equation (7.3.4). The asymptotic solution (7.3.5) can be expressed in the form of the elementary plane wave solution
(7.3.6) φ(x, t) ∼ Re A(xi , t) exp iθ(xi , t) , where Re denotes the real part, and the complex amplitude A(xi , t) and the phase function θ(xi , t) are given by 2 (2π)n/2 F (ki ) iπ ∂ Ω exp − sgn A(xi , t) = , (7.3.7) ∂2 Ω n 1/2 4 ∂ki2 [t det | ∂ki ∂kj |] κ), θ(xi , t) = κ · x − tΩ(κ with the fact that ki (xi , t) is the solution of equation (7.3.4).
(7.3.8)
7.4 Nonlinear Dispersive Waves and Whitham’s Equations
361
The asymptotic solution (7.3.6) has the same form as the elementary plane wave solution, and also represents an oscillatory wavetrain. But, in contrast to the elementary solution, (7.3.6) is not a uniform wavetrain because ki is a function of xi and t. Moreover, the quantities A and ω = Ω(κ) involved in (7.3.6) are no longer constants, but are functions of xi and t. It follows from (7.3.7) combined with stationary phase equation (7.3.4) that
∂Ω ∂ki ∂θ = ki + xi − t = ki (xi , t), (7.3.9) ∂xi ∂ki ∂xi
∂θ ∂Ω ∂ki − (7.3.10) = Ω(κi ) − xi − t = Ω(κi ) = ω(xi , t). ∂t ∂ki ∂t This means that ∂θ/∂xi and −∂θ/∂t still have the significance of a wavenumber and a frequency. They are no longer constants, but functions of xi and t. Thus, ki = ∂θ/∂xi and ω = −∂θ/∂t represent the local wavenumber and the local frequency and are still governed by the dispersion relation (7.2.6). The most remarkable difference between the asymptotic solution (7.3.6) and the elementary plane wave solution (7.2.3) is that the former represents an oscillatory nonuniform wavetrain, but the parameters A, ki , and ω are no longer constants. Indeed, they are slowly varying functions of xi and t in the sense that ΔA A, Δκ κ, and Δω ω over a length scale 2π/κ and a time scale 2π/ω. In other words, the relative changes of these quantities in one wavelength and in one period are very small. This point can readily be verified by using (7.3.4) and (7.3.7). Finally, the present asymptotic analysis reveals two remarkable consequences concerning the dual role of the group velocity. First, a careful consideration of result (7.3.4) reveals that the local wavenumber propagates with the group velocity ∂Ω/∂ki . Second, expression (7.3.7) indicates that |A|2 is an energy-like quantity and also propagates with the group velocity.
7.4 Nonlinear Dispersive Waves and Whitham’s Equations To describe a slowly varying nonlinear and nonuniform oscillatory wavetrain in a dispersive medium, we assume the existence of a one-dimensional solution in the form (7.3.6), so that
φ(x, t) = a(x, t) exp iθ(x, t) + c.c., (7.4.1) where c.c. stands for the complex conjugate and a(x, t) is the complex amplitude given by (7.3.7) with n = 1. The phase function θ(x, t) is given by θ(x, t) = xk(x, t) − tω(x, t),
(7.4.2)
and k, ω, and a are slowly varying functions of space variable x and time t. Because of the slow variations of k and ω, it is reasonable to assume that these quantities still satisfy the dispersion relation of the form
362
7 Nonlinear Dispersive Waves and Whitham’s Equations
(7.4.3)
ω = W (k). Differentiating (7.4.2) with respect to x and t, respectively, we obtain
θx = k + x − tW (k) kx ,
θt = −W (k) + x − tW (k) kt .
(7.4.4) (7.4.5)
In the neighborhood of stationary points defined by W (k) = (x/t) > 0, these equations become θx = k(x, t) and θt = −ω(x, t). (7.4.6ab) These results can be used as a definition of local wavenumber and local frequency of a slowly varying nonlinear wavetrain. In view of (7.4.6ab), relation (7.4.3) gives a nonlinear partial differential equation for the phase θ(x, t) in the form ∂θ ∂θ +W = 0. (7.4.7) ∂t ∂x The solution of this equation determines the geometry of the wave pattern. We eliminate θ from (7.4.6ab) to obtain the equation ∂k ∂ω + = 0. ∂t ∂x
(7.4.8)
This is known as the Whitham equation for the conservation of waves, where k represents the density of waves and ω is the flux of waves. Using the dispersion relation (7.4.3), equation (7.4.8) gives ∂k ∂k + C(k) = 0, ∂t ∂x
(7.4.9)
where C(k) = W (k) is the group velocity. This represents the simplest nonlinear wave (hyperbolic) equation for the propagation of wavenumber k with group velocity C(k). Since equation (7.4.9) is similar to (5.2.11), we can use the analysis of Section 5.2 to find the general solution of (7.4.9) with the initial condition k = f (x) at t = 0. In this case, the solution takes the form k(x, t) = f (ξ),
x = ξ + tF (ξ),
(7.4.10ab)
where F (ξ) = C(f (ξ)). This further confirms the propagation of wavenumber k with group velocity C. A physical interpretation of this kind of solution has already been discussed in Section 5.2. Equations (7.4.9) and (7.4.3) reveal that ω also satisfies the first-order, nonlinear wave (hyperbolic) equation ∂ω ∂ω + W (k) = 0. ∂t ∂x
(7.4.11)
7.4 Nonlinear Dispersive Waves and Whitham’s Equations
363
It follows from equations (7.4.9) and (7.4.11) that both k and ω remain constant on the characteristic curves defined by dx = W (k) = C(k) dt
(7.4.12)
in the (x, t)-plane. Since k or ω is constant on each characteristic, the characteristics are straight lines with slope C(k). The solution for k(x, t) is given by (7.4.10ab). Finally, it follows from the preceding analysis that any constant value of the phase φ propagates according to θ(x, t) = const., and hence, dx (7.4.13) θt + θx = 0, dt which gives, by (7.4.6ab), dx θt ω = − = = c. dt θx k
(7.4.14)
Thus, the phase of the waves propagates with the phase speed c. On the other hand, equation (7.4.9) ensures that the wavenumber k propagates with group velocity C(k) = (dω/dk) = W (k). We next investigate how wave energy propagates in a dispersive medium. We consider the following integral involving the square of the wave amplitude (energy) between any two points x = x1 and x = x2 (0 < x1 < x2 ): x2 x2 2 Q(t) = |A| dx = AA∗ dx (7.4.15) x1 x1 x2 F (k)F ∗ (k) dx, (7.4.16) = 2π t|W (k)| x1 which, due to a change of variable x = tW (k), is
k2
= 2π
F (k)F ∗ (k) dk,
(7.4.17)
k1
where xr = tW (kr ), r = 1, 2. When kr is kept fixed as t varies, Q(t) remains constant so that x2 ∂ dQ d x2 0= = |A|2 dx + |A|22 W (k2 ) − |A|21 W (k1 ). |A|2 dx = dt dt x1 ∂t x1 (7.4.18) In the limit, as x2 − x1 → 0, this result reduces to the first-order, partial differential equation ∂ ∂ |A|2 + W (k)|A|2 = 0. (7.4.19) ∂t ∂t
364
7 Nonlinear Dispersive Waves and Whitham’s Equations
This represents the equation for the conservation of wave energy where |A|2 and |A|2 W (k) are the energy density and energy flux, respectively. It also follows that the energy propagates with group velocity W (k). It has been shown that the wavenumber k also propagates with the group velocity. Thus, the group velocity plays a double role. The preceding analysis reveals another important fact that (7.4.3), (7.4.8), and (7.4.19) constitute a closed set of equations for the three functions k, ω, and A. Indeed, these are the fundamental equations for nonlinear dispersive waves and are known as Whitham’s equations.
7.5 Whitham’s Theory of Nonlinear Dispersive Waves The theory of linear dispersive waves is essentially based on the Fourier superposition principle. However, in nonlinear problems, the superposition principle is no longer applicable to construct a more general solution for a wavetrain. It follows from the asymptotic solution (7.3.6) that the form of the elementary solution φ = A exp(iθ), θ = κ · x − ωt, can still be used to describe nonlinear wavetrains provided that A, ω, and κ are no longer constants but slowly varying functions of x and t corresponding to the slow modulation of the wavetrains. For slowly varying wavetrains, there exists a phase function θ(x, t) so that the local frequency ω(x, t) and the wavenumber vector κ (x, t) = ki are defined in terms of θ by ω(x, t) = −θt ,
ki = θx1 .
(7.5.1ab)
In view of the slow variation of ki and ω, it seems reasonable to assume the linear dispersion relation in the form ω = Ω(ki ). (7.5.2) We note that, for nonlinear wavetrains, Ω may also depend on amplitude and other parameters. Elimination of θ from (7.5.1ab) immediately gives ∂ω ∂ki + = 0, ∂t ∂x ∂ki ∂kj − = 0, ∂xj ∂xi
(7.5.3a) (7.5.3b)
where the former is a relationship between the wavenumbers and frequency. This can be regarded as an equation of continuity for the phase. Substitution of (7.5.2) into (7.5.3a) and use of (7.5.3b) imply ∂ki ∂Ω ∂ki =− , + Cj (k) ∂t ∂xj ∂xi where Cj = Cj (k) is called the three-dimensional group velocity, defined by
(7.5.4)
7.5 Whitham’s Theory of Nonlinear Dispersive Waves
Cj =
∂Ω . ∂kj
365
(7.5.5)
This represents the propagation velocity for the wavenumber ki . An observer traveling with the local group velocity Cj (k) of the wavetrain moves along a path in (xi , t)-space known as a ray. From the existence of the phase function θ(xi , t) and the dispersion relation (7.5.2), it turns out that changes along rays are given by dki ∂Ω , =− dt ∂xi ∂Ω dω = . dt ∂t
(7.5.6a) (7.5.6b)
Equation (7.5.4) can also be written in the characteristic form dki ∂Ω =− , dt ∂xi ∂Ω dxi = Ci . = dt ∂ki
(7.5.7a) (7.5.7b)
These are identical with Hamilton’s equations of motion for a particle with position xi and generalized momentum ki , where Ω plays the role of the Hamiltonian. If Ω is independent of time, that is, if ω = Ω(ki , xi ), then equation (7.5.6b) implies that the frequency ω remains constant for the wavepacket. This is identical with the important fact that, for a dynamical system with a time-independent Hamiltonian, every motion of the system carries a constant value of the Hamiltonian equal to the total energy. Thus, the frequency behaves like energy and the wavenumber like momentum. This remarkable analogy between classical (particle) mechanics and quantum (wave) mechanics leads to the well-known duality exploited in quantum physics. In a uniform medium, ∂Ω/∂xi = 0, equation (7.5.6a) or (7.5.7a) implies that the wavenumber ki is constant on each characteristic, and the characteristics are straight lines in the (xi , t)-space. Also, each ki propagates along the characteristics with the constant group velocity Ci . Further, if the medium is independent of time t, ∂Ω/∂t = 0, and then the frequency ω is constant on each characteristic. In a nonuniform medium, equations (7.5.7a) and (7.5.7b) show significant differences. First, ki is no longer constant, but varies at the rate −∂Ω/∂xi . Second, it propagates along the characteristics with the variable group velocity Ci (ki , xi , t), and third, the characteristics are no longer straight lines. It is interesting to observe that the local dispersion relation (7.5.2) with (7.5.1ab) gives the partial differential equation for phase θ(xi , t) in the form ∂θ ∂θ , xi , t = 0. (7.5.8) +Ω ∂t ∂xi This is the well-known Hamilton–Jacobi equation. The solution of this equation determines the geometry (or the kinematics) of the wavetrain.
366
7 Nonlinear Dispersive Waves and Whitham’s Equations
The preceding kinematic theory provides no indication of changes in the amplitude or energy of a wavetrain or wavepacket. Whitham (1974) initiated further study of the amplitude function (7.3.7) and the related energy density to derive a differential equation for the amplitude and to determine the second role of the group velocity. For the one-dimensional case, the energy density described in (7.3.7) is proportional to |A|2 , so that 2πF (k)F ∗ (k) , (7.5.9) A2 = t|Ω (k)| where F ∗ (k) is the complex conjugate of F (k). The amount of energy Q(t) between two points x1 and x2 is given by x2 g(k)A2 dx, Q(t) =
(7.5.10)
x1
where g(k) is an arbitrary proportionality factor associated with the square of the amplitude and energy. In a new coordinate system moving with the group velocity, that is, along the rays x = C(k)t, equation (7.5.10) reduces to the form x2 Q(t) = 2π g(k)F (k)F ∗ (k) dk, (7.5.11) x1
where Ω (k) > 0 and k1 and k2 are defined by x1 = C(k1 )t and x2 = C(k2 )t, respectively. Using the principle of conservation of energy, that is, stating that the energy between the points x1 and x2 traveling with the group velocities C(k1 ) and C(k2 ) remains invariant, it turns out from (7.5.10) that x2
dQ ∂ = g(k)A2 dx + g(k2 )C(k2 )A2 (x2 , t) dt x1 ∂t − g(k1 )C(k1 )A2 (x1 , t) = 0,
(7.5.12)
which is, in the limit as x2 − x1 → 0,
∂ ∂ g(k)A2 + g(k)C(k)A2 = 0. ∂t ∂t
(7.5.13)
This may be treated as the energy equation. Quantities g(k)A2 and g(k)C(k)A2 represent the energy density and the energy flux, so that they are proportional to |A|2 and C(k)|A|2 , respectively. The flux of energy propagates with the group velocity C(k). Hence, the group velocity has a double role: it is the propagation velocity for the wavenumber k and for the energy g(k)|A|2 . Thus, (7.5.3a) and (7.5.13) are known as Whitham’s conservation equations for nonlinear dispersive waves. The former represents the conservation of wave-number k and the latter is the conservation of energy (or more generally, the conservation of wave action). It is important to observe that, even in a uniform medium
7.6 Whitham’s Averaged Variational Principle
367
(∂Ω/∂xi = 0), equation (7.5.3a) or its equivalent form given by (7.5.4) is the most fundamental hyperbolic equation, even if the original equation (7.2.1) for φ is linear or nonhyperbolic. The conservation equations can be derived more rigorously from a general and extremely powerful approach that is now known as Whitham’s averaged variational principle.
7.6 Whitham’s Averaged Variational Principle In all dynamical problems where the governing equations admit uniform, periodic, wavetrain solutions, it is generally true that the system can be described by the Hamilton variational principle (7.6.1) δ L(ut , uxi , u) dx dt = 0, where L is the Lagrangian and the dependent variable is u = u(xi , t). The Euler–Lagrange equation for (7.6.1) is ∂L2 ∂L1 + − L3 = 0, ∂t ∂xi where L1 =
∂L , ∂ut
L2 =
∂L , ∂uxi
and
L3 =
(7.6.2)
∂L . ∂u
(7.6.3abc)
Equation (7.6.2) is a second-order, partial differential equation for u(xi , t). We assume that this equation has periodic wavetrain solutions in the form u = Φ(θ),
θ = κ · x − ωt,
(7.6.4ab)
where κ and ω are constants and represent the wavenumber vector and the frequency, respectively. Since (7.6.2) is a second-order equation, its solution depends on two arbitrary constants of integration. One is the amplitude a and the other is the phase shift. Omitting the latter constant, it turns out that a solution of (7.6.2) exists provided that the three parameters ω, κ, and a are connected by a dispersion relation D(ω, κ , a) = 0.
(7.6.5)
In linear problems with a wavetrain solution in the form u = Φ(θ) = aeiθ , the dispersion relation (7.6.5) does not involve the amplitude a. For slowly varying dispersive wavetrains, the solution maintains the elementary form u = Φ(θ, a), but ω, κ , and a are no longer constants, so that θ is not a linear function of xi and t. The local wavenumber and local frequency are defined by ki =
∂θ , ∂xi
ω=−
∂θ . ∂t
(7.6.6ab)
368
7 Nonlinear Dispersive Waves and Whitham’s Equations
The quantities ω, ki , and a are slowly varying functions of xi and t corresponding to the slow modulation of the wavetrain. The Whitham averaged Lagrangian over the phase of the integral of the Lagrangian L is defined by 2π 1 L dθ (7.6.7) L(ω, κ , a, x, t) = 2π 0 and is calculated by assuming the uniform periodic solution u = Φ(θ, a) in L. Whitham first formulated the averaged variational principle in the form δ L dx dt = 0, (7.6.8) to derive the equations for ω, κ , and a. It is noted that the dependence of L on x and t reflects possible nonuniformity of the medium supporting the wave motion. In a uniform medium, L is independent of x and t, so that the Whitham function L ≡ L(ω, κ , a). However, in a uniform medium, some additional variables also appear only through their derivatives. They represent potentials whose derivatives are important physical quantities. The Euler equations resulting from the independent variations of δa and δθ in (7.6.8) with L = L(ω, κ , a) are δa : La (ω, κ , a) = 0, ∂ ∂ Lk = 0. δθ : Lω − ∂t ∂xi i
(7.6.9) (7.6.10)
The θ-eliminant of (7.6.6ab) gives the consistency equations ∂ki ∂ω + = 0, ∂t ∂xi
∂ki ∂kj − = 0. ∂xj ∂xi
(7.6.11ab)
Thus, (7.6.9)–(7.6.11ab) represent the Whitham equations for describing the slowly varying wavetrain in a nonuniform medium and constitute a closed set from which the triad ω, κ , and a can be determined. In linear problems, the Lagrangian L, in general, is a quadratic in u and its derivatives. Hence, if Φ(θ) = a cos θ is substituted in (7.6.7), L must always take the form L(ω, κ , a) = D(ω, κ )a2 ,
(7.6.12)
so that the dispersion relation (La = 0) must take the form D(ω, κ ) = 0.
(7.6.13)
We note that the stationary value of L is, in fact, zero for linear problems. In the simple case, L equals the difference between kinetic and potential energy. This proves the well-known principle of equipartition of energy, stating that average potential and kinetic energies are equal.
7.7 Whitham’s Instability Analysis of Water Waves
369
7.7 Whitham’s Instability Analysis of Water Waves Section 7.6 dealt with Whitham’s new remarkable variational approach to the theory of slowly varying, nonlinear, dispersive waves. Based upon Whitham’s ideas and, especially, Whitham’s fundamental dispersion equation (7.6.10), Lighthill (1965, 1967) developed an elegant and remarkably simple result determining whether very gradual—not necessarily small—variations in the properties of a wavetrain are governed by hyperbolic or elliptic partial differential equations. A general account of Lighthill’s work with special reference to the instability of wavetrains on deep water was described by Debnath (1994). This section is devoted to the Whitham instability theory with applications to water waves. According to Whitham’s nonlinear theory, La = 0 gives a dispersion relation that depends on wave amplitude a and has the form κ, a), ω = ω(κ
(7.7.1)
where equations for κ and a are no longer uncoupled and constitute a system of partial differential equations. The first important question is whether these equations are hyperbolic or elliptic. This can be answered by a standard and simple method combined with Whitham’s conservation equations (7.5.3a) and (7.5.13). For moderately small amplitudes, we use the Stokes expansion of ω in terms of k and a2 in the form ω = ω0 (k) + ω2 (k)a2 + · · · ,
(7.7.2)
We substitute this result in (7.5.3a) and (7.5.13), replace ω (k) by its linear value and retain the terms of order a2 to obtain the equations for k and a2 in the form
ω0 (k),
∂k ∂k ∂a2 + ω (k) + ω2 (k)a2 + ω2 (k) = 0, ∂t ∂x ∂x ∂a2 ∂a2 ∂k + ω0 (k) + ω0 (k)a2 = 0. ∂t ∂x ∂x
(7.7.3) (7.7.4)
Neglecting the term O(a2 ), these equations can be rewritten as ∂k ∂a2 ∂k + ω0 + ω2 = 0, ∂t ∂x ∂x ∂a2 ∂a2 ∂k + ω0 (k) + ω0 a2 = 0. ∂t ∂x ∂x
(7.7.5) (7.7.6)
These describe the modulations of a linear dispersive wavetrain and represent a coupled system due to the nonlinear dispersion relation (7.7.2) exhibiting the dependence of ω on both k and a. In matrix form, these equations read ∂k ∂k ω0 ω2 1 0 ∂x ∂t + = 0. (7.7.7) ∂a2 ∂a2 ω0 a2 ω0 0 1 ∂x
∂t
Hence, the eigenvalues λ are the roots of the determinant equation
370
7 Nonlinear Dispersive Waves and Whitham’s Equations
ω0 − λ ω2 = 0, |aij − λbij | = 2 ω0 a ω0 − λ
(7.7.8)
where aij and bij are the coefficient matrices of (7.7.7). This determinant equation gives the characteristic velocities 1/2 dx ≡ C(k) = C0 (k) ± a ω2 (k)ω0 (k) (7.7.9ab) + O a2 , λ= dt where C0 (k) = ω0 (k) is the linear group velocity, and, in general, ω0 (k) = 0 for dispersive waves. The equations are hyperbolic or elliptic depending on whether ω2 (k)ω0 (k) > 0 or < 0. In the hyperbolic case, the characteristics are real, and the double characteristic velocity splits into two separate velocities and provides a generalization of the group velocity of nonlinear dispersive waves. In fact, the characteristic velocities (7.7.9ab) are used to define the nonlinear group velocities. The splitting of the double characteristic velocity into two different velocities is one of the most significant results of the Whitham theory. This means that any initial disturbance of finite extent will eventually split into two separate disturbances. This prediction is radically different from that of the linearized theory, where an initial disturbance may suffer from a distortion due to dependence of the linear group velocity C0 (k) = ω0 (k) on the wavenumber k, but would never split into two. Another significant consequence of nonlinearity in the hyperbolic case is that compressive modulations will suffer from gradual distortion and steepening in the typical hyperbolic manner discussed in Chapter 6. This leads to the multi-valued solutions, and hence, eventually, breaking of waves. In the elliptic case (ω2 , ω0 < 0), the characteristics are imaginary. This leads to ill-posed problems in the theory of nonlinear wave propagation. Any small sinusoidal disturbances in a and k may be represented by solutions of the form exp[ia{x − C(k)t}], where C(k) is given by (7.7.9ab) for the unperturbed values of a and k. Thus, when C(k) is complex, these disturbances will grow exponentially in time, and hence, the periodic wavetrains become definitely unstable. An application of this analysis to the Stokes waves on deep water reveals that the associated dispersion equation is elliptic in this case. For waves on deep water, the dispersion relation is 1 ω = gk 1 + a2 k 2 + O a4 . (7.7.10) 2 √ This result is compared with the Stokes expansion (7.7.2) to give ω0 (k) = gk and √ ω2 (k) = 12 k 2 gk. Hence, ω0 ω2 < 0, the velocities (7.7.9ab) are complex, and the Stokes waves on deep water are definitely unstable. The instability of deep water waves came as a real surprise to researchers in the field in view of the very long and controversial history of the subject. The question of instability went unrecognized for a long period of time, even though special efforts have been made to prove the existence of a permanent shape for periodic water waves for all amplitudes less than the critical value at which the waves assume a sharp-crested form. However, Lighthill’s
7.8 Whitham’s Equation: Peaking and Breaking of Waves
371
(1965) theoretical investigation into the elliptic case and Benjamin and Feir’s (1967) theoretical and experimental findings have provided conclusive evidence of the instability of Stokes waves on deep water. For more details of nonlinear dispersive wave phenomena, the reader is referred to Debnath (1994).
7.8 Whitham’s Equation: Peaking and Breaking of Waves In 1865, Rankine conjectured that there exists a wave of extreme height. In a moving frame of reference, the Euler equations are Galilean invariant, and the Bernoulli equation on the free surface takes the form 1 |∇φ|2 + gρη = E 2 which expresses the conservation of local energy. The first term represents the kinetic energy of the fluid, where φ is the velocity potential, and the second term is the potential energy due to gravity. For the wave of extreme height, E = gρηmax , where ηmax is the maximum height of the elevation. Thus, at the maximum height of the fluid, the velocity is zero, and there is a stagnation point in the flow. Rankine had conjecture that the free surface formed a cusp at the peak, that is, the tangent lines to the free surface are vertical. Stokes (1847) also suggested that such a wave of extreme height exists and the angle subtended at the peak is 120◦ . Toland (1978) proved the existence of a wave of extreme height and showed also that if the singularity at the peak is not a cusp, then Stokes’ conjecture about the value of the subtended angle is true. Subsequently, Amick et al. (1982) proved rigorously that the singularity at the peak is not a cusp. However, the full Euler equations of motion exhibit singularities, and also there is a limiting amplitude of the periodic waves, or of the solitary wave. This means that the KdV approximation can only be valid for sufficiently small amplitudes. But the KdV equation has smooth solitary and cnoidal waves of arbitrary amplitude. In his pioneering work on nonlinear water waves, Whitham (1974) observed that the neglect of dispersion in the nonlinear shallow water equations leads to the development of multi-valued solutions with a vertical slope, and hence, eventually breaking occurs. It seems clear that the third derivative dispersive term in the KdV equation produces the periodic and solitary waves which are not found in the shallow water theory. However, the KdV equation cannot describe the observed symmetrical peaking of the crests with a finite angle. On the other hand, the Stokes waves include full effects of dispersion, but they are limited to small amplitude, and describe neither the solitary waves nor the peaking phenomenon. Although both breaking and peaking are without doubt involved in the governing equations of the exact potential theory, Whitham (1974) developed a mathematical equation that can include all these phenomena. It has been shown earlier that the breaking of shallow water waves is governed by the nonlinear equation
372
7 Nonlinear Dispersive Waves and Whitham’s Equations
ηt + c0 ηx + dηηx = 0,
d = 3c0 (2h0 )−1 .
(7.8.1)
On the other hand, the linear equation corresponding to a general linear dispersion relation ω = c(k) (7.8.2) k is given by the integrodifferential equation in the form ∞ ηt + K(x − s)ηs (s, t) ds = 0, (7.8.3) −∞
where the kernel K is given by the inverse Fourier transform of c(k): ∞
1 eikx c(k) dk. K(x) = F −1 c(k) = 2π −∞
(7.8.4)
Whitham combined the above ideas to formulate a new class of nonlinear nonlocal equations ∞
ηt + dηηx +
−∞
K(x − s)ηs (s, t) ds = 0.
(7.8.5)
This is well known as the Whitham equation which can, indeed, describe symmetric waves that propagate without change of shape and peak at a critical height, as well as asymmetric waves that invariably break. Once a wave breaks, it usually continues to travel in the form of a bore as observed in tidal waves. The weak bores have a smooth but oscillatory structure, whereas the strong bores have a structure similar to turbulence with no coherent oscillations. Since the region where waves break is a zone of high energy dissipation, it is natural to include a second derivative dissipative term in the KdV equation to obtain (7.8.6) ηt + c0 ηx + dηηx + μηxxx − νηxx = 0, where μ = 16 c0 h20 . This is known as the KdV–Burgers equation which also arises in nonlinear acoustics for fluids with gas bubbles (Karpman 1975a). It is convenient to rewrite the Whitham equation (7.8.5) in the form ∞ K(x − s)us (s, t) ds = 0, (7.8.7) ut + uux + −∞
and assume that u(x, t) is the classical solution of (7.8.7) with the property u → 0 as |x| → ∞, and the kernel K (or convolution operator) is symmetric. ˜ by We next define an integral operator K ∞ ˜ (Ku)(x, t) = K(x − s)u(s, t) ds, (7.8.8) −∞
˜ v = so that it is bounded and self-adjoint in the Hilbert space L2 (R), i.e., Ku, ˜ ˜ u, Kv defined in Debnath and Mikusinski (1999). Also, this operator K commutes ∂ ˜ x = ∂x K. ˜ , i.e., K∂ with ∂x ≡ ∂x
7.8 Whitham’s Equation: Peaking and Breaking of Waves
373
We next show that the Whitham equation admits solitary traveling wave solutions in the form u(x, t) = f (x − U t), where U is the velocity of the wave. Substituting this into (7.8.7) gives ˜ 0 f = 0, (7.8.9) −U f + f f + K ˜ 0 is associated with the kernel K0 (x) = where the operator K It is easy to verify that, for v ∈ L2 (R) ∩ C(R), 2 ˜ 0 v = −ν 2 v. ∂x − ν 2 K
π 4
exp(−ν|x|), ν =
π 2.
(7.8.10)
Integrating (7.8.9) with respect to x gives 1 ˜ 0 f − A = 0, −U f + f 2 + K 2
(7.8.11)
where A is an integrating constant. Applying the operator (∂x2 − ν 2 ) to (7.8.11) and using (7.8.10) yields 2 1 2 2 + ν 2 (f − A) = 0. Uf − f ∂x − ν (7.8.12) 2 Multiplying (7.8.12) by ∂x (U f − 12 f 2 ) = (U − f )f and integrating with respect to x gives 2 2 1 (U − f )2 f 2 − ν 2 U f − f 2 − U f 2 + f 3 2 3 1 − 2Aν 2 U f − f 2 = B, 2
(7.8.13)
where B is also a constant of integration. We consider the special case where A = B = 0 so that (7.8.13) can be put into the form (U − f )2 f 2 = ν 2 f 2 Q(f ), (7.8.14) where Q(f ) ≡
2 1 2 f + U 2 − U. f − U− 4 3
(7.8.15)
Periodic solutions correspond to oscillations of f between two simple zeros of Q(f ). If 1 < U < 34 , there are two simple zeros f0 and f1 where 0 < f0 < f < f1 . Consequently, (7.8.14) can be rewritten as
dx df
2 =
(U − f )2 ≡ F 2 (f ) ν 2 f 2 Q(f )
(say),
(7.8.16)
which can be solved in 0 ≤ f ≤ f0 . Integrating (7.8.16) gives (x − x0 )2 = h2 (f ),
dh = F (f ), df
(7.8.17)
374
7 Nonlinear Dispersive Waves and Whitham’s Equations
where x0 is a constant of integration and f = f0 corresponds to x = x0 . Equa) is constant because 0 < f0 < f < f1 , and tion (7.8.16) shows that the sign of ( dx df tends to ±∞ as f → 0+ and f0 → f0 − 0. It turns out that the solution f represents a solitary wave. It is well known that there are infinitely many polynomial conservation laws for the shallow water equation, the KdV equation, the Benjamin–Ono equation, and other integrable equations. These equations are special cases of the Whitham equation. Recently, Benguria and Depassier (1989) have shown that these equations are the only representatives of the Whitham equation that possess this property. We next consider a few conservation laws for the Whitham equation. The first conservation law is ∞ u(x, t) dx = const., (7.8.18) I1 (u) = −∞
that is, d d I1 (u) = I1 (u0 ) = 0, u(x, 0) = u0 (x). (7.8.19ab) dt dt This is obtained directly from: ∞ ∞ ∞ d I1 (u) = ut (x, t) dx = − K(x − s)us (s, t) ds dx uux + dt −∞ −∞ −∞ ∞ ∞ 1 ∞ d 2 u dx − =− dx K(x − s)us (s, t) ds 2 −∞ dx −∞ −∞ ∞ ∞ ∞ 1 = − u2 − us (s, t) ds K(x − s) dx 2 −∞ −∞ −∞ ∞ = −c0 (0) u(s, t) −∞ = 0,
where
∞
K(z) dz.
c(0) =
(7.8.20)
−∞
Thus it follows that I1 (u) = I1 (u0 ) = const.
(7.8.21)
This represents the conservation of mass. The second conservation law I2 (u) = const. can be derived by multiplying (7.8.7) by u and then integrating the result with respect to x by parts so that ∞ 3 ∞ 1 d ∞ 2 1 d u ˜ dx u (x, t) dx = − ∂x u∂x Ku I2 (u) = dx − 2 dt 2 dt −∞ 3 −∞ −∞ ˜ = u, K∂ ˜ x u = u, ∂x Ku ˜ = ∂x u, Ku ˜ = 0. = − u, ∂x Ku This means that I˙2 (u) = 0, giving I2 (u) = I2 (u0 ) = const.
7.8 Whitham’s Equation: Peaking and Breaking of Waves
The third conserved quantity I3 (u) is given by ∞ 1 3 ˜ u + uKu dx. I3 (u) = −∞ 3 It turns out that d I3 (t) = dt = = = =
375
(7.8.22)
∞
˜ + u, Ku ˜ t u2 ut dx + ut , Ku −∞ ∞ ˜ − ˜ x u u2 dx uux + K∂ 2 ut , Ku −∞ 4 ∞ 1 2 u ˜ , Ku ˜ ˜ xu u + Ku −2 ∂x − − u2 , K∂ 2 4 −∞ 2 2 ˜ xu ˜ + 2 Ku, ˜ ∂x Ku ˜ − u , K∂ u , ∂x Ku ∞ ∂ ˜ 2 Ku dx = 0. −∞ ∂x
This quantity I3 (u) represents the energy. Since it is independent of time t, the Whitham equation (7.8.7) can be written in the Hamiltonian form ∂ ∂H ∂u =0 (7.8.23) + ∂t ∂x ∂x where the Hamiltonian is H = 12 I3 (u). We next make some important comments on the possible extension of a class of nonlinear equations. First, if we eliminate the requirement that the kernel K(x) is even, the Whitham equation (7.8.7) can not only describe the complete dispersion, but also dissipation processes. This extends the class of nonlinear equation under consideration. Second, another natural extension is to generalize (7.8.7) to include nonlocal equations with an arbitrary pseudodifferential operator so that (7.8.7) can be replaced by the following equation: ut + uux + K(u) = 0, where the operator K(u) is given by ∞ 1 K(u) = eikx K(k) u(k, t) dk, 2π −∞ where u (k, t) is the Fourier transform of u(x, t) defined by ∞
u (k, t) = u(x, t) = eikx u(x, t) dx.
(7.8.24)
(7.8.25)
(7.8.26)
−∞
The function K(k) introduced to define the operator K(u) is called the symbol of the operator. The dispersion relation c(k) and the symbol K(k) are related by the result
376
7 Nonlinear Dispersive Waves and Whitham’s Equations
K(k) = ikc(k).
(7.8.27)
This approach reveals a new fact that the Whitham equation represents, as particular cases, many nonlinear equations which are of special interest for physical applications. Examples of such equations include the Kawahara (1972) equation ut + uux + auxxx − uxxxxx = 0,
(7.8.28)
where K(k) = −i(ak 3 +k5 ). This equation describes propagation of signals in electric transmission lines, long waves under ice cover in liquids of finite depth (Ilichev and Marchenko 1989), and water waves with surface tension (Zufira 1987). Another example is the Kuramoto–Sivashinsky equation 1 ut + u2x + uxx + auxxxx = 0 2
(7.8.29)
which arises in the theory of combustion to model a flame front and also in twodimensional turbulence (Kuramoto 1984, and Novick-Cohen and Sivashinsky 1986). Finally, we close this section by adding some specific comments on peaking and breaking of waves described by the Whitham equation (7.8.7) with the initial condition (7.8.30) u(x, 0) = u0 (x). It has been shown by Whitham (1974) that if K(x) = α exp(−β|x|), where α, β > 0 are constants, the crests of the waves of limiting amplitude have an angle of 110◦ , which is very close to the Stokes angle. Naumkin and Shishmarev (1994) proved a theorem which states that the wave remains peaked for all time as long as a solution exists provided the kernel K(x) in (7.8.7) is continuous and absolutely integrable. On the other hand, waves described by (7.8.7) and (7.8.30) break in a finite time T provided the kernel K(x) in (7.8.7) is regular as shown by Seliger (1968). The first study of the Cauchy problem (7.8.7) and (7.8.30) for the Whitham equation with a singular kernel K(x) was initiated by Naumkin and Shishmarev (1994). They showed that a wave that is sufficiently steep initially at t = 0 breaks in a finite time T provided the kernel K(x) is monotonically increasing in the neighborhood of the singular point x = 0 and has a singularity of order |x|−α , 0 < α < 1 as x → 0, and also if the kernel K(x) is monotonically decreasing near x = 0 and has a singularity of order |x|−α , 0 < α < 23 . For a monotonically decreasing kernel, the integral term in the Whitham equation (7.8.7) represents dissipative effects, whereas the monotonically increasing kernel corresponds to a process of energy pumping. Making reference to Chapter 7 of Naumkin and Shishmarev (1994) and Amick et al. (1989), we simply state the asymptotic behavior of a solution u(x, t) of the Cauchy problem for the nonlocal nonlinear Whitham equation, as t → ∞, ut + u2x + K(u) = 0, x ∈ R, t ≥ 0, (7.8.31) u(x, 0) = u0 (x), where K(u) is defined by (7.8.25) and (7.8.26).
7.9 Exercises
The asymptotic behavior of u(x, t) as t → ∞ is given by ∞ δ 1 −λt −1/δ cos(kξ)e−ωk dk + O e−λt t− δ −μ , u(x, t) ∼ Ae t
377
(7.8.32)
0
where ξ = xt−1/δ ≥ 0, the constant A is expressed through the symbol K(k) and the initial function u0 (x), and all other quantities are defined in Naumkin and Shishmarev (1994, p. 180).
7.9 Exercises 1. (a) For the plane wave solutions φ(x, t) = A exp{i(kx − ωt)} of the real partial differential equation ∂φ ∂ + iP −i φ = 0, ∂t ∂x where P is an odd function, show that the dispersion relation is ω = P (k). (b) For a wave packet described by φ(x, t) = Re[A(x, t)ψ(x, t)], where A(x, t) represents a slowly varying complex amplitude function, and where ψ(x, t) = exp[i(k0 x − ω0 t)] denotes the basic carrier wave of frequency ω0 = P (k0 ) for a given wavenumber k0 = 0, show that the amplitude A(x, t) satisfies the equation 1 i At + P (k0 )Ax + P (k0 )Axx = 0, 2 where P (k) has a Taylor series expansion about k = k0 . 2. The dispersion relation (Whitham 1967a) for water waves in an inviscid fluid of arbitrary depth h0 is ω = ω0 (k) + Ω2 (k)
k2 E + O E2 , ρc0
1
where ω0 (k) = (gk tanh kh0 ) 2 , E = 12 gρa2 , and (9T0 − 10T02 + 9) 1 1 (4C0 − c0 )2 , − Ω2 (k) = 1+ 8T03 kh0 4 (gh0 − C02 ) where T0 = tanh(kh0 ), c0 = ω0k(k) , and C0 = 0 velocities given by (7.7.9ab) are C(k) = C0 (k) ±
dω0 dk0 .
ω0 (k)Ω2 (k)Ek 2 ρc0
Show that the characteristic
12 ,
ω0 (k) < 0.
(a) Explain the significance of the solution for Ω2 (k) > 0 or < 0. (b) If the critical value for instability is determined by the numerical value of (kh0 ) for which Ω2 (k) = 0, show that equations are hyperbolic or elliptic according to whether kh0 < 1.363 or > 1.363.
378
7 Nonlinear Dispersive Waves and Whitham’s Equations
3. If the Whitham averaged variational principle is given by δ L(−θt , θx ) dt dx = 0, where ω = −θt and k = θx , show that the Euler equation is ∂ ∂ Lω − Lk = 0. ∂t ∂x Hence, or otherwise, show that the phase function θ satisfies the second-order, quasi-linear equation Lωω θtt − 2Lωk θtx + Lkk θxx = 0. 4. The Whitham equations for the slow modulation of the wave amplitude a and the wavenumber k in the case of two-dimensional deep water waves are 2 ∂ ∂k ∂ω a ∂ a2 + = 0 and C + = 0, ∂t ω0 ∂x ω0 ∂t ∂x √ where ω0 = gk and C = 2ωg 0 is the group velocity. Using Chu and Mei’s (1971) dispersion relation
a 1 2 2 a k + ω = ω0 1 + ε2 div 2ω0 a , 2 ω0 tt derive the following equations for the phase function φ(x, t), where we have used Chu and Mei’s notation W = −2φx : a2 ∂a2 ∂ 2 ∂ axx ∂ ∂φ + a φx = 0 and − 2 + −φ2x + + = 0. ∂t ∂x ∂t ∂x ∂x 4 16a Show that the second equation can be integrated with respect to x to obtain 1 1 axx = 0. φt + φ2x − a2 − 2 8 32a Using Ψ = a exp(4iφ), show that Ψ (x, t) satisfies the nonlinear Schrödinger equation 1 1 iΨt + Ψxx + Ψ |Ψ |2 = 0. 8 2 5. Show that Whitham’s equation (7.8.7) possess the conserved quantity ∞ I4 (u) = xu(x, t) dx. −∞
6. Show that the concentric KdV equation (see Johnson 1997) ut +
u − 6uux + uxxx = 0 2t
has the following conservation laws:
7.9 Exercises
379
∞ √ (a) −∞ tu dx = const., ∞ (b) −∞ tu2 dx = const., ∞ √ (c) −∞ ( txu + 6t3/2 u2 ) dx = const. 7. Using the shallow water equations (2.7.67)–(2.7.70), derive the conservation law 1 1 1 1 m3 + hm1 + m4 + hm2 + m21 + h3 = 0, 3 3 2 3 t x where
h
un dz,
mn = 0
(see Johnson 1997, Benney 1974, and Miura 1974). 8. Consider the KP equation ut − 6uux + uxxx + 3vy = 0,
uy = vx .
Under suitable decay conditions, show that the second equation produces three conservation laws: ∞ (a) −∞ u dx = const., ∞ (b) −∞ u dy = const., ∞ (c) −∞ v dx = const. 9. Show that the Boussinesq equations (see Hirota 1973b) ut + hx − 3 h2 x + hxxx = 0, ht + u x = 0 havethe conserved quantities ∞ (a) −∞ u dt, ∞ (b) −∞ uh dx, and ∞ √ (c) −∞ ( txu + 6t3/2 u2 ) dx. 10. Introducing the averaged Hamiltonian H by 2π 1 ∂L H= · ut − L dθ = ωJ − L, 2π 0 ∂ut where the wave action J is defined by 2π ∂L ∂L 1 , J= · ut dθ = 2πω 0 ∂ut ∂ω show that the Whitham equations ∂k ∂ω + =0 ∂t ∂x
and
∂ ∂L ∂ ∂L − =0 ∂t ∂ω ∂x ∂k
can be expressed in the averaged Hamiltonian form ∂ ∂H ∂J ∂ ∂H ∂k + = 0 and + = 0. ∂t ∂x ∂J ∂t ∂x ∂k
8 Nonlinear Diffusion–Reaction Phenomena
The profound study of nature is the most fertile source of mathematical discoveries. Joseph Fourier The research worker, in his efforts to express the fundamental laws of Nature in mathematical form, should strive mainly for mathematical beauty. He should take simplicity into consideration in a subordinate way to beauty. . . . It often happens that the requirements of simplicity and beauty are the same, but where they clash the latter must take precedence. Paul Dirac
8.1 Introduction Many physical phenomena are described by the interaction of convection and diffusion and also by the interaction of diffusion and reaction. From a physical point of view, the convection–diffusion process and the diffusion–reaction process are quite fundamental in describing a wide variety of problems in physical, chemical, biological, and engineering sciences. Some nonlinear partial differential equations that model these processes provide many new insights into the question of interaction of nonlinearity and diffusion. It is well known that the Burgers equation is a simple nonlinear model equation representing phenomena described by a balance between convection and diffusion. The Fisher equation is another simple nonlinear model equation which arises in a wide variety of problems involving diffusion and reaction. To understand these physical processes, this chapter is devoted to the study of both Burgers and Fisher equations and their different kinds of solutions with physical significance. Special attention is given to diffusive wave solutions and traveling wave solutions of the Burgers and Fisher equations. In addition to the standard mathematical methods used for solving the Burgers and Fisher equations, similarity methods L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_8,
382
8 Nonlinear Diffusion–Reaction Phenomena
are developed to find the similarity solutions of both linear and nonlinear diffusion equations with examples of applications.
8.2 Burgers Equation and the Plane Wave Solution We recall the differential form of the nonlinear conservation equation ∂ρ ∂q + = 0. ∂t ∂x
(8.2.1)
To investigate the nature of the discontinuous solution or shock waves, we assume a functional relation q = Q(ρ) and allow a jump discontinuity for ρ and q. In many physical problems of interest, it would be a better approximation to assume that q is a function of the density gradient ρx as well as ρ. A simple model is to take (8.2.2) q = Q(ρ) − νρx , where ν is a positive constant. Substituting (8.2.2) into (8.2.1), we obtain the nonlinear diffusion equation (8.2.3) ρt + c(ρ)ρx = νρxx , where c(ρ) = Q (ρ). We multiply (8.2.3) by c (ρ) to obtain ct + ccx = νc (ρ)ρxx = ν cxx − c (ρ)ρ2x .
(8.2.4)
If Q(ρ) is a quadratic function in ρ, then c(ρ) is linear in ρ, and c (ρ) = 0. Consequently, (8.2.4) becomes ct + ccx = νcxx . (8.2.5) As a simple model of turbulence, c is replaced by the fluid velocity field u(x, t) to obtain the well-known Burgers equation as ut + uux = νuxx ,
(8.2.6)
where ν is the kinematic viscosity. Thus, the Burgers equation is a balance between time evolution, nonlinearity, and diffusion. This is the simplest nonlinear model equation for diffusive waves in fluid dynamics. Burgers (1948) first developed this equation primarily to shed light on the study of turbulence described by the interaction of the two opposite effects of convection and diffusion. However, turbulence is more complex in the sense that it is both three dimensional and statistically random in nature. Equation (8.2.6) arises in many physical problems including one-dimensional turbulence (where this equation had its origin), sound waves in a viscous medium, shock waves in a viscous medium, waves in fluid-filled viscous elastic tubes, and magnetohydrodynamic waves in a medium with finite electrical conductivity. We note that (8.2.6) is parabolic, whereas (8.2.6)
8.2 Burgers Equation and the Plane Wave Solution
383
with ν = 0 is hyperbolic. More importantly, the properties of the solution of the parabolic equation are significantly different than those of the hyperbolic equation. We first solve (8.2.6) for two simple cases: (i) a linearized Burgers equation and (ii) an equation with the linearized convective term cux , where c is a constant. In the first case, the linearized Burgers equation ut = νuxx ,
−∞ < x < ∞, t > 0,
with the initial conditions at t = 0
u(x, 0) =
u1 u2
if x < 0, if x > 0,
(8.2.7)
(8.2.8ab)
can readily be solved by applying the Fourier transform with respect to x. The final solution of (8.2.7) with (8.2.8ab) for t > 0 is 1 x 1 √ . (8.2.9) u(x, t) = (u1 + u2 ) − (u1 − u2 ) erf 2 2 2 νt This shows that the presence of the diffusion term νuxx is to smooth out the ini1 tial distribution like (νt)− 2 . The solution (8.2.9) tends to constant values u1 , as x → −∞, and u2 , as x → +∞. The absence of the diffusion term in (8.2.6) leads to gradual nonlinear steepening, and eventually breaking. Indeed, equation (8.2.6) combines the two opposite effects of nonlinearity and diffusion. In the absence of the diffusion term in (8.2.6), the resulting equation reduces to the firstorder, nonlinear wave equation which admits a progressively distorted wave profile as a solution. Eventually this solution develops a discontinuity as a shock wave. In the second case, equation (8.2.6) reduces to the linear parabolic equation ut + cux = νuxx . We seek a plane wave solution of (8.2.10) in the form u(x, t) = a exp i(kx − ωt) ,
(8.2.10)
(8.2.11)
where Im ω = −νk < 0, since ν > 0. Thus, the solution (8.2.11) becomes 2
2 u(x, t) = ae−νk t exp ik(x − ct) .
(8.2.12)
This represents a diffusive wave with wavenumbers k and phase velocity c. The amplitude of the wave decays exponentially with time t, and the decay time t0 = (νk 2 )−1 becomes smaller as k increases with fixed ν. Thus, the waves of smaller wavelengths decay faster than the waves of longer wavelengths. On the other hand, for a fixed wavenumber k, the decay time decreases as ν increases so that waves of a given wavelength attenuate faster in a medium with a larger ν. This quantity ν may be regarded as a measure of diffusion. Finally, after a sufficiently long time (t t0 ), only disturbances of long wavelength will survive, whereas all waves of
384
8 Nonlinear Diffusion–Reaction Phenomena
short wavelength will decay very rapidly. For ν < 0, the solution (8.2.12) tends to infinity as t → ∞, and hence, it becomes unstable.
8.3 Traveling Wave Solutions and Shock-Wave Structure To investigate the effects of nonlinear steepening and diffusion, we seek a traveling wave solution of the Burgers equation (8.2.6) in the form u(x, t) = u(ξ),
ξ = x − U t,
(8.3.1ab)
where the wave speed U is to be determined and u(ξ) represents the wave form with the property that it tends asymptotically to constant values u1 , as ξ → −∞, and u2 , as ξ → +∞. We assume that u1 > u2 . Substituting (8.3.1ab) into the Burgers equation (8.2.6) gives the ordinary differential equation −U u (ξ) + uu (ξ) − νu (ξ) = 0. Integrating this equation yields 1 −U u(ξ) + u2 − νu (ξ) = A, 2 where A is a constant of integration, or equivalently, u (ξ) =
1 2 u − 2U u − 2A . 2ν
(8.3.2)
Clearly, this suggests that u1 and u2 are the roots of the quadratic equation u2 − 2U u − 2A = 0.
(8.3.3)
Hence, U and A are determined from the sum and the product of the roots u1 and u2 of (8.3.3), and therefore, U=
1 (u1 + u2 ), 2
1 A = − u1 u2 . 2
(8.3.4ab)
Thus, the wave speed is the average of the two speeds of asymptotic states at infinity. Equation (8.3.2) can now be written as 2ν
du = u2 − 2U u − 2A = (u − u1 )(u − u2 ). dξ
Integrating this equation gives ξ du u1 − u du =− . = log 2ν (u1 − u)(u − u2 ) (u1 − u2 ) (u − u2 )
(8.3.5)
8.3 Traveling Wave Solutions and Shock-Wave Structure
385
Fig. 8.1 Traveling wave solution u(ξ), ξ = x − U t.
This leads to the solution for u(ξ) in the form u(ξ) =
ξ )(u1 − u2 )] u1 + u2 exp[( 2ν
ξ 1 + exp[( 2ν )(u1 − u2 )] (u1 − u2 ) . = u2 + ξ 1 + exp[( 2ν )(u1 − u2 )]
(8.3.6) (8.3.7)
Another useful expression for u can be written from (8.3.6) in the form ξ u1 + u2 exp[( 2ν )(u1 − u2 )] 1 1 − (u1 − u2 ) (u1 + u2 ) + ξ 2 2 1 + exp[( 2ν )(u1 − u2 )]
1 1 ξ (u1 − u2 ) . = (u1 + u2 ) − (u1 − u2 ) tanh (8.3.8) 2 2 4ν
u(ξ) =
As u1 > u2 , the wave profile u(ξ) decreases monotonically with ξ from the constant value u1 , as ξ → −∞, to the constant value u2 , as ξ → +∞, as shown in Figure 8.1. At ξ = 0, u = 12 (u1 +u2 ). The shape of the waveform (8.3.8) is significantly affected by the diffusion coefficient ν. This means that the presence of diffusion processes prevents the gradual distortion of the wave profile, and so, it does not break. On the other hand, if the diffusion term is absent (ν = 0) in (8.2.6), the wave profile will suffer from gradual distortion and steepening, and hence, it would definitely break with development of a shock. Shock waves are formed as a result of a balance between the steepening effect of the convective (nonlinear) term and the smoothing effect of the linear diffusive terms in the equation of motion. The tendency to steepening has been demonstrated for plane waves by many authors, including Riemann (1858), who first introduced the most general approach. The upshot of the above analysis is that the convection and diffusion terms in the Burgers equation exhibit opposite effects. The former introduces a sharp discontinuity in the solution profile, whereas the latter tends to diffuse (spread out) the discontinuity into a smooth profile. In view of this property, the solution is called
386
8 Nonlinear Diffusion–Reaction Phenomena
the diffusive wave. In the context of fluid flow, ν represents the kinematic viscosity which measures the viscous dissipation. ξ (u1 − u2 )], Multiplying both numerator and denominator of (8.3.6) by exp[− 2ν we write the solution (8.3.6) in the form u(ξ) =
ξ )(u1 − u2 )] u2 + u1 exp[−( 2ν ξ 1 + exp[−( 2ν )(u1 − u2 )]
.
(8.3.9)
The exponential factor in this solution indicates the existence of a thin transition layer of thickness δ of the order ν/(u1 − u2 ). This thickness δ can be referred to as the shock thickness, which tends to zero as ν → 0 for fixed u1 and u2 . Also, δ increases as u1 → u2 for a fixed ν. If δ is small compared with other typical length scales of the problem, the rapid shock transition can satisfactorily be approximated by a discontinuity. Thus, in the limit as ν → 0, we might expect that solutions of (8.2.6) tend to solutions of the nonlinear equation ut + uux = 0
(8.3.10)
together with discontinuous shock waves satisfying the jump condition
1 2 u1 − u22 = U (u1 − u2 ), 2
(8.3.11)
where U is the shock speed given by U=
1 (u1 + u2 ). 2
(8.3.12)
This is in complete agreement with the above analysis.
8.4 The Exact Solution of the Burgers Equation We solve the initial-value problem for the Burgers equation ut + uux = νuxx , u(x, 0) = F (x),
x ∈ R, t > 0, x ∈ R.
(8.4.1) (8.4.2)
Special attention will be given to small values of ν or to large values of Reynolds numbers. Hopf (1950) and Cole (1951) independently discovered a transformation that reduces the Burgers equation to a linear diffusion equation. First, we write (8.4.1) in a form similar to a conservation law, 1 2 u − νux ut + = 0. (8.4.3) 2 x
8.4 The Exact Solution of the Burgers Equation
387
This can be regarded as the compatibility condition for a function ψ to exist, such that u = ψx and 1 νux − u2 = ψt . 2
(8.4.4a) (8.4.4b)
We substitute the value of u from (8.4.4a) in (8.4.4b) to obtain 1 νψxx − ψx2 = ψt . 2
(8.4.5)
Next, we introduce ψ = −2ν log φ so that u = ψx = −2ν
φx . φ
(8.4.6)
This is called the Cole–Hopf transformation, which, by differentiating, gives 2 2ν φt φx φxx and ψt = −2ν . ψxx = 2ν − φ φ φ Consequently, (8.4.5) reduces to the linear diffusion equation (8.4.7)
φt = νφxx .
Many solutions of this equation are well known in the literature. We substitute the given solution for φ to find solutions of the Burgers equation. We now solve equation (8.4.7) subject to the initial condition φ(x, 0) = Φ(x),
x ∈ R.
(8.4.8a)
This can be written in terms of the initial value u(x, 0) = F (x) by using (8.4.6): F (x) = u(x, 0) = −2ν
φx (x, 0) . φ(x, 0)
(8.4.8b)
Integrating this result gives x 1 F (α) dα . φ(x, 0) = Φ(x) = exp − 2ν 0
(8.4.9)
The Fourier transform method or the joint Fourier–Laplace transform technique can be used to solve the linear initial-value problem (8.4.7), (8.4.8a), and hence, the standard solution of this problem is
∞ (x − ζ)2 1 dζ, (8.4.10) Φ(ζ) exp − φ(x, t) = √ 4νt 2 πνt −∞ where Φ(ζ) is given by (8.4.9). We then substitute the value of Φ(ζ) to rewrite (8.4.10) in the convenient form
388
8 Nonlinear Diffusion–Reaction Phenomena
1 φ(x, t) = √ 2 πνt where
f dζ, exp − 2ν −∞
∞
ζ
F (α) dα +
f (ζ, x, t) = 0
Thus, 1 φx (x, t) = − √ 4ν πνt
∞ −∞
x−ζ t
(x − ζ)2 . 2t
f exp − dζ. 2ν
(8.4.11)
(8.4.12)
(8.4.13)
Therefore, the exact solution of the Burgers initial-value problem is obtained from (8.4.6) in the form ∞ x−ζ f ( t ) exp(− 2ν ) dζ . (8.4.14) u(x, t) = −∞ ∞ f exp(− 2ν ) dζ −∞ This is clearly single-valued and continuous for all values of t. The physical interpretation of this exact solution can hardly be given unless a suitable simple form of F (x) is specified. In many problems, an exact evaluation of the integrals involved in (8.4.14) is almost a formidable task. It is then necessary to resort to asymptotic methods. We next consider the following example to investigate the formation of discontinuities or shock waves. Example 8.4.1. Find the solution of the Burgers initial-value problem with physical significance for the discontinuous data (8.4.15)
F (x) = Aδ(x)H(x),
where A is a constant, δ(x) is the Dirac delta function, and H(x) is the Heaviside unit step function. To find the solution, first we calculate ζ (x−ζ)2 − A if ζ < 0, (x − ζ)2 2t δ(α) dα + f (ζ, x, t) = A = (x−ζ) 2 2t if ζ > 0. 0+ 2t
Thus, the integral in the numerator of (8.4.14) is 0 ∞ x−ζ x−ζ f A (x − ζ)2 exp − dζ = exp − dζ t 2ν t 2ν 4νt −∞ −∞ ∞ (x − ζ)2 x−ζ exp − dζ + t 4νt 0
x2 = 2ν eR − 1 exp − , 4νt √ A which is obtained by substituting (x − ζ)/2 νt = α and writing ( 2ν ) = R which can be interpreted as the Reynolds number (Whitham 1974). Clearly, by small ν, we mean large R (R 1), and the large ν corresponds to small R (R 1).
8.4 The Exact Solution of the Burgers Equation
389
Similarly, the integral in the denominator of (8.4.14) gives ∞ √ √ R
f x dζ = 2 νt π + e − 1 erfc √ . exp − 2ν 2 νt −∞ So the final solution u(x, t) is obtained from (8.4.14) in the form u(x, t) =
x2 (eR − 1) exp(− 4νt ) ν √ . t √π + (eR − 1) π erfc( √x ) 2 2 νt
(8.4.16)
Thus, this solution takes the similarity form ν u(x, t) = g(η, R), (8.4.17) t √ where η = (x/2 νt) is the similarity variable. Two limiting cases (i) R → 0 and (ii) R → ∞ are of interest, and they are considered below. Case (i) In the limit as R → 0(ν → ∞), the effect of diffusion would be more significant than that of nonlinearity. So, in this case, we can approximate eR ∼ 1 + R and erfc( 2√xνt ) → 0 in (8.4.16) to obtain the final solution x2 A exp − . u(x, t) ∼ √ 4νt 2 πνt
(8.4.18)
This is the fundamental (or source) solution of the linear diffusion equation. At any time t, of u(x, t) decreases inversely √ the velocity field is Gaussian. The peak height √ √ with νt, whereas the width of the peak (x ∼ νt) increases with νt. These are remarkable features of diffusion phenomena. Case (ii) In this case, R → ∞(ν → 0). So, the nonlinear effect would dominate over diffusion, and hence, discontinuity √ as a shock is√expected√to develop. We introduce the similarity variable η = x/ 2At so that x/ 4νt = η R, and then, rewrite the solution (8.4.16) as
ν (eR − 1) exp(−Rη 2 ) √ , (8.4.19) u(x, t) = √ t √π + (eR − 1)( π ) erfc(η R) 2 which has the similarity form u(x, t) =
A 2Rt
12 g(η, R).
Since R → ∞, eR − 1 can be replaced by eR , hence, (8.4.19) gives
ν exp{R(1 − η 2 )} √ for all η, u(x, t) = √ t √π + ( π ) exp(R) erfc(η R) 2 ∼ 0 as R → ∞ for η < 0 and η > 1.
(8.4.20)
(8.4.21)
390
8 Nonlinear Diffusion–Reaction Phenomena
When R → ∞ and 0 < η < 1, we use the asymptotic result
√
1 exp −η 2 R erfc η R ∼ √ η πR to obtain the solution from (8.4.21) as
1
2A 2 η √ u(x, t) = t 1 + 2η πR exp{−R(1 − η 2 )} 12 2A x ∼η as R → ∞. = t t It turns out that the final asymptotic solution, as R → ∞, is √ x if 0 < x < 2At, t u(x, t) ∼ 0 otherwise.
(8.4.22) (8.4.23)
(8.4.24)
√ This result represents a shock wave at x = 2At, and the shock speed is U = dx dt = 1 A 12 2 , and hence, the ) . The solution u(x, t) has a jump from zero to xt = ( 2A ) ( 2t t shock condition is satisfied. Whitham (1974) investigated the structure of the shock wave solution for large values of R. He found two transition layers: one at η = 1 and the other (weaker) layer at η = 0. For large but finite R, (8.4.22) indicates a rapid transition from
exponentially small values in η > 1 to u ∼ 2A t η in η < 1. In the transition layer η ≈ 1, solution (8.4.22) can be approximated by √ −1 2A 1 + 2 πR exp 2R(η − 1) u(x, t) ∼ . (8.4.25) t
This shows that this transition layer has thickness O(R−1 ). On the other hand, the second layer at η = 0 does smooth out the discontinuity in the derivative between u ∼ 0 in η < 0 to u ∼ xt in 0 < η < 1. It follows from (8.4.21) that this weaker layer occurs for η = O( √1R ), and hence, (8.4.22) assumes the approximate form ν exp(−Rη 2 ) √ . · (8.4.26) u(x, t) ∼ 2 πt erfc(η R) t u is plotted against This solution for large R is drawn in Figure 8.2, where 2ν x √ η = 2νt . This represents a triangular wave solution of the Burgers equation. As R → ∞, the shock layer at x = 0 becomes a discontinuity of ux .
8.5 The Asymptotic Behavior of the Burgers Solution
391
Fig. 8.2 Triangular wave solution of the Burgers equation.
8.5 The Asymptotic Behavior of the Burgers Solution We use the stationary phase approximation to examine the asymptotic nature of the Burgers solution (8.4.14). We consider any typical integral u(x, t) =
b
F (k) exp itθ(k) dk,
(8.5.1)
a
where F (k) is a given function and θ(k) = ( xt )k − ω(k). Integral (8.5.1) can be approximated asymptotically, as t → ∞ for fixed x, by u(x, t) ∼ F (k1 )
2π t|θ (k1 )|
12
π exp i tθ(k1 ) + sgn θ (k1 ) , 4
(8.5.2)
where k1 is a stationary point determined by the solution of the equation θ (k) =
x − ω (k) = 0, t
a < k1 < b
(8.5.3)
and θ (k1 ) = 0. According to this result, the significant contribution to integrals involved in (8.4.14) comes from stationary points for fixed x and t, that is, from the roots of the equation x−ζ ∂f = 0. (8.5.4) = F (ζ) − ∂ζ t Suppose that ζ = ξ(x, t) is a solution of (8.5.4) representing a stationary point. Invoking the stationary phase formula (8.5.2), the integrals in (8.4.14) have the asymptotic representation as ν → 0, in the form,
∞ −∞
1 f (ξ) x−ζ f x−ζ 4πν 2 exp − exp − dζ ∼ , t 2ν t |f (ξ)| 2ν
1 ∞ 4πν 2 f f (ξ) dζ ∼ . exp − exp − 2ν |f (ξ)| 2ν −∞
392
8 Nonlinear Diffusion–Reaction Phenomena
Therefore, the final asymptotic solution is u(x, t) ∼
x−ξ , t
where ξ(x, t) satisfies equation (8.5.4). In other words, the solution assumes the asymptotic form u = F (ξ), ξ = x − tF (ξ).
(8.5.5)
(8.5.6ab)
This is identical with the solution of the Burgers equation without the diffusion term (ν = 0). Here, the stationary point ξ corresponds to the characteristic variable in the context of the first-order, quasi-linear equation. As discussed in Chapter 5, the wave profile described by (8.5.6ab) suffers from gradual distortion and steepening, leads to a multi-valued solution after a sufficiently long time, and, eventually, it breaks with the development of discontinuity as a shock wave. When this state is reached, mathematically, there will be two stationary points of (8.5.4), and then, some modification is required to complete the asymptotic analysis. This leads to Whitham’s geometrical rule of equal area which is equivalent to the shock condition. We shall not pursue this analysis further, and refer to Whitham (1974) and Burgers (1974) for a detailed discussion.
8.6 The N -Wave Solution To find an N -wave solution of the Burgers equation, we begin with the source solution of the linear diffusion equation (8.4.7) of the form x2 τ exp − , (8.6.1) φ(x, t) = 1 + t 4νt where τ is a constant. Substituting (8.6.1) into (8.4.6) gives the solution of the Burgers equation as τ x2 x t exp(− 4νt ) u(x, t) = · , x2 t 1 + τt exp(− 4νt ) 2 −1 t x x . = 1+ exp t τ 4νt
(8.6.2) (8.6.3)
For any time t > 0, this solution is shown in Figure 8.3 and has the N -shaped form. Because of this particular shape of the wave profile, it is known as the N -wave solution of the Burgers equation. At t = t0 > 0, the profile may be taken as the initial profile. Such N waves are observed in many physical situations including problems governed by the cylindrical or spherical Burgers equation.
8.6 The N -Wave Solution
393
Fig. 8.3 N -wave solution of the Burgers equation.
The area under the negative phase of the wave profile is the same as that over the positive phase of the profile. So, the area under the positive phase of the wave profile is given by ∞ ∞ τ . (8.6.4) u(x, t) dx = −2ν log φ(x, t) 0 = 2ν log 1 + t 0 We denote the right-hand side of (8.6.4) in the initial time t = t0 by A and then introduce a Reynolds number R by τ A , (8.6.5) R= = log 1 + 2ν t0 so that tτ0 = (eR − 1), and solution (8.6.3) reduces to u(x, t) =
2 −1
−1 x R x t R 1+ e −1 exp . t t0 2At
When R 1, eR − 1 ∼ eR , so that (8.6.6) reduces to the form −1 t x x2 u(x, t) ∼ 1+ exp −R 1 − . t t0 2At
(8.6.6)
(8.6.7)
In the limit, as R → ∞ with fixed t, (8.6.7) gives the shock-wave solution of the Burgers equation in the form √ x if|x| < 2At, √ (8.6.8) u(x, t) ∼ t 0 if |x| > 2At. In the limit as t → ∞, for fixed ν and τ , solution (8.6.2) takes the form 12 τ x x2 . u(x, t) ∼ exp − t t 4νt This corresponds to the dipole solution of the linear diffusion equation.
(8.6.9)
394
8 Nonlinear Diffusion–Reaction Phenomena
Finally, in the limit as R → 0, eR − 1 ∼ R, and result (8.6.6) gives 2 t0 x R xR exp − . u(x, t) ∼ t t 2At
(8.6.10)
This is identical with (8.6.9), as expected, because R → 0 corresponds to ν → ∞.
8.7 Burgers Initial- and Boundary-Value Problem We solve Burgers equation (8.4.1) in 0 < x < l, t > 0 with the following initial and boundary conditions: πx , 0 ≤ x ≤ l, (8.7.1) u(x, 0) = u0 sin l u(0, t) = u(l, t) = 0, t > 0. (8.7.2) It follows from the Cole–Hopf transformation, that is, from (8.4.9), that x πα u0 sin φ(x, 0) = exp − dα 2ν l 0 πx u0 l 1 − cos , = exp − 2πν l
(8.7.3)
so that boundary conditions (8.7.2) are satisfied. The standard solution of the linear diffusion equation (8.4.7) is given by 2 2 ∞ nπx n π νt cos , (8.7.4) φ(x, t) = a0 + an exp − l2 l n=1 where πx u0 l 1 − cos dx exp − 2πν l 0 lu0 u0 l I0 = exp − 2πν 2πν
1 a0 = l
(8.7.5)
and lu0 πx nπx exp − 1 − cos cos dx 2πν l l 0 u0 l lu0 = 2 exp − In , 2πν 2πν
2 an = l
(8.7.6)
where the preceding integrals are evaluated by using standard integrals from Abramowitz and Stegun (1972), and I0 (x) and In (x) are the modified Bessel functions of the first kind.
8.7 Burgers Initial- and Boundary-Value Problem
395
Thus, the solution of the Burgers initial-boundary problem is given by u(x, t) =
4πν l
∞
2
2
u0 l nIn ( 2πν ) exp(− n πl2 νt ) sin( nπx ) l . (8.7.7) ∞ u0 l u0 l n2 π 2 νt I0 ( 2πν ) + 2 n=1 In ( 2πν ) exp(− l2 ) cos( nπx l ) n=1
u0 l ) cos( πx At t = 0, the value of the denominator of (8.7.7) is exp[( 2πν l )]. We use this result combined with (8.7.1) and the value of φx (x, 0) from (8.7.3) and (8.7.4) to verify that the solution (8.7.7) satisfies the initial condition (8.7.1). We note here that the quantity ( uν0 l ) = R represents the Reynolds number. The solution of the linear diffusion equation (8.4.7), subject to the initial and boundary data (8.7.1), (8.7.2), is given by π 2 νt πx sin . (8.7.8) φ(x, t) = u0 exp − 2 l l
The following conclusions are in order. First, the nonlinear solution (8.7.7) contains an infinite set of higher harmonics with decreasing amplitudes, whereas the linear solution (8.7.8) contains only the fundamental harmonic. Second, the former solution depends on the Reynolds number R rather than the initial amplitude u0 . We next use the asymptotic expansion of the modified Bessel function 1 R R ∼√ , as R → ∞, exp In π π 2R to find the asymptotic solution for u(x, t), as R → ∞, u(x, t) ∼
4πν l
∞
1+
n2 π 2 νt ) sin( nπx n=1 exp(− l2 l ) . ∞ 2 2 2 n=1 exp(− n πl2 νt ) cos( nπx ) l
(8.7.9)
It is interesting to point out that this asymptotic solution is identical with the exact ), solution. This series solution is extremely rapidly convergent for all values of ( νt l2 and hence, it is easy to calculate the wave form. When νt 1, the term with n = 1 2 l in (8.7.9) dominates the series, and hence, we obtain πx π 2 νt 4πν sin exp − 2 . (8.7.10) u(x, t) ∼ l l l This represents a sinusoidal wave form with an exponentially decaying attenuation. Thus, the ultimate decay of the solution is dominated by diffusion. Another form of the solution (8.7.9) can be derived by using the Jacobi theta function defined by ϑ3 (X, T ) = 1 + 2
∞ n=1
where X =
πx 2l
and T =
νπt l2 ,
so that
exp −πn2 T cos(2nX),
396
8 Nonlinear Diffusion–Reaction Phenomena
−1 ∞ ∂ νnπ 2 t n log ϑ3 (X, T ) = 2 (−1) sinh sin(2nX). ∂X l2 n=1 In view of these results, solution (8.7.9) becomes u(x, t) =
2πν l
−1 νnπ 2 t nπx . (−1)n+1 sinh sin l2 l n=1
∞
(8.7.11)
This is another form of the solution (8.7.9). As suggested by Cole (1951), this solution is approximately equal to l l−x l−x u(x, t) ∼ tanh − . (8.7.12) t 2νt l This does not depend on the initial amplitude. Physically, the initial sinusoidal profile for large R suffers from a nonlinear steepening effect near x = l because of the higher harmonics generated by convection. However, this wave form steepening may be prevented by the effects of diffusion. As time t becomes large, this effect spreads in the entire wave profile, leading to exponential decay of the harmonics according to (8.7.11). Finally, for a detailed discussion of the solution of the spatially periodic initialvalue problem with general initial conditions, we refer to a paper by Walsh (1969).
8.8 Fisher Equation and Diffusion–Reaction Process Fisher (1936) first introduced a nonlinear evolution equation to investigate the wave propagation of an advantageous gene in a population. His equation also describes the logistic growth–diffusion process and has the form u , (8.8.1) ut − νuxx = ku 1 − κ where ν(> 0) is a diffusion constant, k(> 0) is the linear growth rate, and κ(> 0) is the carrying capacity of the environment. The term f (u) = ku(1 − uκ ) represents a nonlinear growth rate which is proportional to u for small u, but decreases as u increases, and vanishes when u = κ. It corresponds to the growth of a population u when there is a limit κ on the size of the population that the habitat can support; if u > κ, then f (u) < 0, so the population decreases whenever u is greater than the limiting value κ. This interpretation suggests that the habitat can support a certain maximum population so that 0 ≤ u(x, 0) ≤ κ for x ∈ R.
(8.8.2)
In recent years, the Fisher equation (8.8.1) has been used as a basis for a wide variety of models for the spatial spread of genes in a population and for chemical
8.8 Fisher Equation and Diffusion–Reaction Process
397
wave propagation. It is pertinent to mention recent work on gene-culture waves of advance by Aoki (1987), on the propagation of chemical waves by Arnold et al. (1987), and on the spread of early farming in Europe by Ammerman and CavalliSforva (1971, 1983). It also represents a model equation for the evolution of a neutron population in a nuclear reactor (Canosa 1969, 1973). The Fisher equation (8.8.1) is a particular case of a general model equation, called the nonlinear reaction–diffusion equation, which can be obtained by introducing the net growth rate f (x, t, u) so that it takes the form ut − νuxx = f (x, t, u),
x ∈ R, t > 0.
(8.8.3)
The term f is also referred to as a source or reaction term and it represents the birth–death process in an ecological context. This equation arises in many physical, biological, and chemical problems involving diffusion and nonlinear growth. For example, if a chemically reacting substance is diffusing through a medium, then its concentration u(x, t) satisfies (8.8.3), where f represents the rate of increase of the substance due to the chemical reaction. The temperature distribution u(x, t) satisfies (8.8.3) when a chemical reaction generates heat at a rate depending on the temperature. Other problems described by (8.8.3) include the spread of animal or plant populations and the evolution of neutron populations in a nuclear reactor, where f represents the net growth rate. We study the Fisher equation as a nonlinear model for a physical system involving linear diffusion and nonlinear growth. It is convenient to introduce the nondimensional quantities x∗ , t∗ , u∗ defined by x∗ =
12 k x, ν
t∗ = kt,
u∗ = κ−1 u,
(8.8.4)
where νk , k −1 , and κ represent the length scale, time scale, and population scale, respectively. Using (8.8.4) and dropping the asterisks, equation (8.8.1) takes the nondimensional form ut − uxx = u(1 − u),
x ∈ R, t > 0.
(8.8.5)
In the spatially homogeneous problem, the stationary states are u = 0 and u = 1, which represent unstable and stable solutions, respectively. It is then appropriate to look for traveling wave solutions of (8.8.5) for which 0 ≤ u ≤ 1. We seek all solutions of (8.8.5) subject to (8.8.2) with κ = 1 such that all derivatives of u vanish as |x| → ∞, and lim u(x, t) = 1
x→−∞
and
lim u(x, t) = 0,
x→∞
t ≥ 0.
(8.8.6)
Physically, the first condition implies that the population has its maximum value as x → −∞, and the second condition represents zero population as x → +∞. In their study of a nonlinear system, with applications to combustion and wave propagation in biology and chemistry, Kolmogorov et al. (1937) proved that, for all
398
8 Nonlinear Diffusion–Reaction Phenomena
initial data of the type 0 ≤ u(x, 0) ≤ 1, x ∈ R, the solution of (8.8.5) is also bounded for all x and t, that is, 0 ≤ u(x, t) ≤ 1,
x ∈ R, t > 0.
They also showed that, for the two sets of discontinuous initial data, 1 if x < 0, (i) u(x, 0) = 0 if x > 0, ⎧ ⎪ if x < a, ⎨1 (ii) u(x, 0) = f (x) if a < x < b, ⎪ ⎩ 0 if x > b, where f (x) is an arbitrary function, the solution of (8.8.5) evolves, as t → ∞, into a shock-like traveling wave satisfying condition (8.8.6) and propagates to the right with minimum characteristic velocity cmin = 2. It is important to note that equation (8.8.5) is invariant under the transformation x → −x. So, it is sufficient to consider waves propagating to the right only. Kolmogorov et al. (1937) suggested one of the best-known model equations with dissipation in the form ut − κuxx = F (u), (8.8.7) where F (u) is an arbitrary function of u. This is known as the KPP equation and it describes phenomena such as combustion, evolution of genes, and propagation of a nerve pulse in biological systems. The generalized KPP equation is given by ut − u2 + K(u) = 0,
(8.8.8)
when the linear pseudodifferential operator κ is defined by (7.8.25) and (7.8.26). The equation (8.8.8) corresponds to the symbol K(k) = 1 + k 2 . The Cauchy condition for (8.8.8) is (8.8.9) u(x, 0) = u0 (x), where u0 (x) ∈ L(R), which is required to ensure the stable solution (u → 0) of the Cauchy problem for (8.8.8), (8.9.9) as t → ∞. Kolmogorov et al. (1937) proposed a new approach to studying the asymptotic behavior of the solution of the nonlocal KPP equation as t → ∞. This approach is essentially based on application of the explicit form of Green’s function for the linear KPP equation. They considered the step-decaying problem for the KPP equation, that is, the Cauchy problem with the initial data u0 (x) → a± as |x| → ∞, where a+ , a− are constants with a+ = a− . The work of Kolmogorov et al. (1937) led to a large number of studies on the asymptotic solutions as t → ∞ of the Cauchy problem for the KPP equation (see also Bramson 1983). Indeed, the uniform asymptotic behavior of u(x, t) with respect to ξ, as t →
8.9 Traveling Wave Solutions and Stability Analysis
∞, is given by A u(x, t) ∼ √ e−t t
∞
2 e−k cos(kξ) dk + O e−t t−1 ,
399
(8.8.10)
0
|x| ) ∈ R, and the constant A is expressed in terms of the symbol where ξ = ( √ t K(k) and the initial function u0 (k). For details, the reader is referred to Chapter 6 in Naumkin and Shishmarev (1994).
8.9 Traveling Wave Solutions and Stability Analysis The Fisher equation (8.8.5) is one of the prototype equations which admits traveling wave solutions. We seek a traveling wave solution of (8.8.5) in the form u(x, t) = u(ξ),
ξ = x − ct,
(8.9.1)
where the wave speed c is to be determined and the waveform u(ξ) satisfies the boundary data (8.8.6), as ξ → ∓∞. We substitute (8.9.1) into (8.8.5) to obtain the nonlinear, ordinary differential equation u (ξ) + cu (ξ) + u(ξ) − u2 (ξ) = 0. Writing
du dξ
(8.9.2)
= v, equation (8.9.2) gives two first-order equations dv = u(u − 1) − cv, dξ
du = 0 · u + v, dξ
(8.9.3ab)
or equivalently,
u(u − 1) − cv dv = . (8.9.4) du v This system (8.9.3ab) has a simple interpretation in the Poincaré phase plane. The singular points of this system are the solutions of the equations v = 0,
u(u − 1) − cv = 0.
(8.9.5ab)
Thus, there are two singular points (u, v) = (0, 0) and (1, 0) which represent the steady states. We follow the standard phase plane analysis to examine the nature of the nonlinear autonomous system given by du = p(u, v), dξ
dv = q(u, v). dξ
The matrix associated with this system at the critical point (u0 , v0 ) is pu (u0 , v0 ) pv (u0 , v0 ) . qu (u0 , v0 ) qv (u0 , v0 )
400
8 Nonlinear Diffusion–Reaction Phenomena
Fig. 8.4 Phase plane trajectories for c > 2.
In the present problem, the matrix A at (0, 0) is 0 1 A(0, 0) = . −1 −c The matrix at (1, 0) is
A(1, 0) =
0 1
1 . −c
The eigenvalues λ of the matrix A at (0, 0) are the roots of the equation 0 − λ 1 = 0. −1 −(c + λ) This equation gives the eigenvalues λ=−
1 c ∓ c2 − 4 . 2
(8.9.6)
The eigenvalues are real, distinct, and of the same sign if the discriminant D = c2 − 4 > 0. According to the theory of dynamical systems, the origin is a stable node for c ≥ cmin = 2, and when c = cmin = 2, it is a degenerate node. The phase plane trajectories of equation (8.9.2) for the traveling wave front solution for c > 2 are shown in Fig. 8.4. There is a unique separatrix joining the stable node (0, 0) with the saddle point (1, 0). If c2 < 4, the eigenvalues are complex with a negative real part, and hence, the curve is a stable spiral, that is, u(ξ) oscillates in the vicinity of the origin. On the other hand, the eigenvalues λ of the matrix A at (1, 0) are the roots of the equation 0 − λ 1 = 0. 1 −c − λ
8.9 Traveling Wave Solutions and Stability Analysis
401
Fig. 8.5 Traveling wave solution.
This equation gives two eigenvalues λ=
1 −c ± c2 + 4 . 2
(8.9.7)
These eigenvalues are real and of opposite sign, and hence, (1, 0) is a saddle point. Finally, the traveling wave solution u(ξ) of the Fisher equation for c > 2, satisfying the boundary conditions (8.8.6), as ξ → ∓∞, is a monotonically decreasing function of ξ when its first derivative vanishes as ξ → ∓∞. Clearly, this corresponds to the separatrix joining the singular points (1, 0) and (0, 0), as shown in Figure 8.4. A typical traveling wave solution u(ξ) for c ≥ 2 is shown in Figure 8.5. On the other hand, if c < 2, there are traveling wave solutions, but they are physically unrealistic since u(ξ) < 0 for some ξ because, in this case, u spirals around the origin. Further, u → 0 at the leading edge with diminishing oscillations about u = 0. We examine the stability of the traveling wave solution of (8.8.5) in the form u(x, t) = U (ξ),
ξ = x − ct,
(8.9.8)
where c is the wave speed. Such a solution is said to be asymptotically stable if a small perturbation imposed on the system at time t = 0 decays to zero, as t → ∞. We introduce a coordinate frame moving with the wave speed c, that is, ξ = x−ct, t = t, so that (8.8.5) reduces to the form (8.9.9) ut − cuξ = uξξ + u(1 − u). We look for a perturbed solution of (8.9.9) in the form u(ξ, t) = U (ξ) + εv(ξ, t),
(8.9.10)
where ε (0 < ε 1) is a small parameter and the second term represents a small perturbation from the given basic solution U (ξ). The perturbed state v(ξ, t) is assumed to have compact support in the moving frame, that is, v(ξ, t) = 0 for |ξ| ≥ a for some finite a > 0. We substitute (8.9.10) into (8.9.9) to find the equation for the perturbed quantity, v, to order ε,
402
8 Nonlinear Diffusion–Reaction Phenomena
vt − cvξ = vξξ + (1 − 2U )v.
(8.9.11)
We next seek solutions of this equation in the form v(ξ, t) = V (ξ) exp(−λt).
(8.9.12)
Substituting this solution into (8.9.11) gives the following linear eigenvalue problem: V (ξ) + cV (ξ) + (λ + 1 − 2U )V = 0,
(8.9.13)
where the growth factor λ is considered an eigenvalue. Since the original perturbed state has compact support, we may take the boundary condition on V (ξ) as V (±a) = 0. According to the general theory of eigenvalue problems, the perturbation will grow in time if the eigenvalues λ are negative, and hence, the system will become unstable. On the other hand, if the eigenvalues λ are positive, then the perturbation will decay to zero, as t → ∞, and hence, the system is asymptotically stable. To reduce the problem in the standard form, we introduce a new transformation 1 (8.9.14) V (ξ) = w(ξ) exp − cξ 2 so that the preceding eigenvalue problem reduces to the form w + [λ − q(ξ)]w = 0, w(−l) = w(l) = 0,
(8.9.15ab)
where
c2 − (1 − 2U ) ≥ 2U (ξ) > 0 for c ≥ 2. (8.9.16) 2 This is a standard eigenvalue problem, and all its eigenvalues are real and positive provided c ≥ 2. This means that all small perturbations of finite extent decay exponentially to zero, as t → ∞. We conclude that the traveling wave solution of the Fisher equation is asymptotically stable. However, the present perturbation analysis is not completely general. The general problem has been studied by several authors including Hoppensteadt (1975) and Larson (1978). q(ξ) =
8.10 Perturbation Solutions of the Fisher Equation The existence of the traveling wave solution of the Fisher equation was established by using a geometric argument. However, it has not been possible to determine the exact or approximate representation of the solution. We now employ the standard perturbation method to find the asymptotic solution of the boundary-value problem for u(ξ), which satisfies the differential system u + cu + u(1 − u) = 0, u(−∞) = 1, where c ≥ 2.
−∞ < ξ < ∞,
(8.10.1)
u(+∞) = 0,
(8.10.2)
8.10 Perturbation Solutions of the Fisher Equation
403
Since equation (8.10.1) is autonomous, the solution is translation invariant. We choose the value u(ξ) at ξ = 0 to be any number in the range of u, and hence, we take u(0) = 12 , which is needed to solve the problem. We introduce a new independent √ variable z = ξc = εξ, where ε = c−2 is a small parameter, to transform (8.10.1) and the boundary conditions into the form εu (z) + u (z) + u − u2 = 0, −∞ < z < ∞, 1 u(−∞) = 1, u(0) = , u(+∞) = 0. 2 We seek a perturbation series expansion for u(z) in powers of ε as u(z, ε) = u0 (z) + εu1 (z) + ε2 u2 (z) + · · · ,
(8.10.3) (8.10.4)
(8.10.5)
where u0 (z), u1 (z), and u2 (z) are to be determined. We substitute (8.10.5) into (8.10.3) and set the coefficients of various powers of ε to be zero to obtain u0 + u0 − u20 = 0, u0 (−∞) = 1,
(8.10.6) 1 u0 (0) = , 2
u0 (+∞) = 0,
(8.10.7)
and u1 + u1 (1 − 2u0 ) + u0 = 0, u1 (−∞) = u1 (0) = u1 (+∞) = 0.
(8.10.8) (8.10.9)
The general solution of (8.10.6) is given by
−1 u0 (z) = 1 − Aez , where A = −1, since u0 (0) = 12 . Thus, the solution reduces to the form
−1 . u0 (z) = 1 + ez We then solve (8.10.8), which becomes z ez e −1 u1 + = 0. u1 − z e +1 (1 + ez )2
(8.10.10)
(8.10.11)
This can be solved directly by elementary methods, and the solution is given by ez 4ez u1 (z) = log . (8.10.12) (1 + ez )2 (1 + ez )2 Thus, the asymptotic solution of the original problem is given by
ε2 ez 4ez z −1 + log u(z) ∼ 1 + e + O ε2 , z 2 z 2 (1 + e ) (1 + e )
(8.10.13)
where z = ( ξc ) = 1c (x − ct). This represents an asymptotic traveling wave solution for c ≥ 2. The present problem is not a singular perturbation problem because u and du dz tend to finite values in the limit, as ε → 0. Equation (8.10.3) has a uniform limit and the regular perturbation method gives an accurate solution in the whole domain.
404
8 Nonlinear Diffusion–Reaction Phenomena
8.11 Method of Similarity Solutions of Diffusion Equations Birkhoff (1950) first recognized that Boltzmann’s method of solving the diffusion equation with a concentration-dependent diffusion coefficient is based on the algebraic symmetry of the equation, and special solutions of this equation can be obtained by solving a related ordinary differential equation. Such solutions are called similarity solutions because they are geometrically similar. He also suggested that the algebraic symmetry of the partial differential equations can be used to find similarity solutions of other partial differential equations by solving associated ordinary differential equations. Thus, the method of similarity solutions became very successful in dealing with the determination of a group of transformations under which a given partial differential equation is invariant. The simplifying feature of this method is that a similarity transformation of the form u(x, t) = tp v(η), η = xt−q , can be found which can then be used effectively to reduce the partial differential equation to an ordinary differential equation with η as the independent variable. The resulting ordinary differential equation is relatively easy to solve. In practice, this method is simple and useful in finding solutions of both linear and nonlinear partial differential equations. We illustrate the method of similarity solutions by examples of applications. Example 8.11.1 (Similarity Solutions of Linear Diffusion Equation). We consider the classical linear diffusion equation with constant diffusion coefficient κ in the form ut = κuxx ,
0 ≤ x < ∞, t > 0,
(8.11.1)
subject to the following boundary and initial conditions: u(0, t) = 1,
u(x, t) → 0
as x → ∞ for t > 0,
u(x, 0) = 0
for 0 ≤ x < ∞.
(8.11.2ab) (8.11.3)
We introduce a one-parameter set of stretching transformations in the (x, t, u)space defined by t˜ = aβ t, u ˜ = aγ u, (8.11.4) x ˜ = aα x, under which equation (8.11.1) is invariant, where a is a real parameter which belongs to an open interval I containing a = 1, and α, β, and γ are the fixed constants. Usually the set of transformations in the (x, t, u)-space is denoted by Ta , and we write the set explicitly as Ta : R3 → R3 for each a ∈ I. The set of all such transformations {Ta } form a Lie group on R3 with an identity element T1 . This can be seen as follows. The set {Ta } obeys the composition (multiplication) law Ta Tb x = Tab x,
for all a, b ∈ I.
(8.11.5)
This law is commutative because Tab = Tba . The associative law is also satisfied since (8.11.6) Ta (Tb Tc ) = Ta (Tbc ) = Tabc = (Tab )Tc = (Ta Tb )Tc .
8.11 Method of Similarity Solutions of Diffusion Equations
405
In view of the fact that for all a = 0,
T1 Ta = T1a = Ta1 = Ta
(8.11.7)
where T1 represents the identity transformation. Finally, we obtain (8.11.8)
Ta Ta−1 = T1 = Ta−1 Ta . This shows that the inverse of Ta is Ta−1 . Clearly, u ˜t˜ = aγ−β ut ,
u ˜x˜ = aγ−α ux ,
u ˜x˜x˜ = aγ−2α uxx ,
(8.11.9)
ux˜x˜ = aγ−β ut − κaγ−2α uxx = aγ−β (ut − κuxx ), u ˜t˜ − κ˜
(8.11.10)
and hence,
provided β = 2α. Hence, equation (8.11.1) is invariant under the transformation x ˜ = aα x,
t˜ = a2α t,
u ˜ = aγ u,
(8.11.11)
for any choice of α and γ. The quantities η = xt−q ,
v = tp u(x, t),
(8.11.12ab)
are invariant under Ta provided p = −(γ/β) and q = (α/β). Thus, the invariants of the transformations are given by v(η) = ut−γ/β = ut−γ/2α ,
η = xt−α/β = xt− 2 . 1
(8.11.13ab)
Substituting (8.11.13ab) into the original equations (8.11.1)–(8.11.3) gives an ordinary differential equation of the form 1 γ v(η) = 0. κv (η) + ηv (η) − 2 2α
(8.11.14)
The transformed data are then given by u(0, t) = tγ/2α v(0) = 1 and v(∞) = 0.
(8.11.15a) (8.11.15b)
To make (8.11.15a) independent of t, we require that γ = 0. Consequently, (8.11.14)–(8.11.15a), (8.11.15b) become η v (η) + (8.11.16) v (η) = 0, 2κ v(0) = 1 and
v(∞) = 0.
(8.11.17ab)
406
8 Nonlinear Diffusion–Reaction Phenomena
Thus, equation (8.11.1) admits the set of transformations (8.11.13ab), which reduces the partial differential equation (8.11.1) to the ordinary differential equation (8.11.14). Result (8.11.13ab) is called a similarity transformation, and the new independent variable η is called a similarity variable. Integrating (8.11.16) yields the general solution 2 η ξ exp − dξ, (8.11.18) v(η) = A + B 4κ 0 where A and B are integration constants to be determined by using (8.11.17ab). It turns out that A = 1 and B = − √1πκ . Thus, the solution takes the form 2 η 1 ξ √ v(η) = 1 − dξ exp − πκ 0 4κ √ x/ 4κt
2 exp −α2 dα = 1− √ π 0 x x = erfc √ , = 1 − erf √ 4κt 4κt
(8.11.19)
where erf(z) and erfc(z) are the standard error and complementary error functions, respectively. The solution (8.11.19) is identical with the known solution (1.9.24) which can be obtained by other methods. The present method of solution seems to be simple and powerful, and hence, could be used for other partial differential equations. Example 8.11.2 (Similarity Solution of the Boltzmann Nonlinear Diffusion Problem). We apply the similarity method to solve the nonlinear diffusion equation ut = (uux )x ,
x ∈ R, t > 0,
(8.11.20)
with the boundary and initial conditions u(±∞, t) = 0, t > 0, u(x, 0) = δ(x), x ∈ R, and
(8.11.21) (8.11.22)
∞
u(x, t) dx = 1
for all t > 0.
(8.11.23)
−∞
We use the set of transformations (8.11.4) in (8.11.20) to obtain ˜2x˜ − u ˜u ˜x˜x˜ = aγ−β ut − a2(γ−α) u2x − a2(γ−α) uuxx u ˜t˜ − u
= aγ−β ut − u2x − uuxx ,
(8.11.24)
provided γ = 2α − β. Hence the given equation (8.11.20) is invariant under the set of transformations
8.11 Method of Similarity Solutions of Diffusion Equations
x ˜ = aα x,
t˜ = aβ t,
u ˜ = a2α−β u.
407
(8.11.25)
The invariants of the set of transformations (8.11.25) are given by v = tp u(x, t),
η = xt−q ,
(8.11.26ab)
provided p = (β − 2α)/β and q = α/β. Therefore, equation (8.11.20) admits the similarity transformation given by u = t−p v(η) = t
2α−β β
v(η),
η = xt− β . α
To determine α and β, we use (8.11.23) to obtain ∞ ∞ 2α−β 3α−β t β v(η) dx = t β v(η) dη = 1, −∞
(8.11.27ab)
(8.11.28)
−∞
which is independent of t provided (α/β) = 13 . Thus, the similarity transformations (8.11.27ab) assume the form u(x, t) = t− 3 v(η), 1
η = xt− 3 . 1
(8.11.29ab)
Putting these results into the given equation (8.11.20) gives an ordinary differential equation for v(η) (8.11.30) 3(vv ) + ηv + v = 0. Integrating this equation once with respect to η gives 3vv + ηv = A,
(8.11.31)
where A is a constant of integration to be determined from the fact that the solution is symmetric about η = 0, that is, ux (0, t) = 0, and hence, v (0) = t2/3 ux (0, t) = 0. Thus, the constant A must be zero; hence, the resulting equation (8.11.31) can be integrated with the boundary condition (8.11.21), that is, v(±∞) = 0, to obtain the solution 1 2 (c − η 2 ) if |η| < c, (8.11.32) v(η) = 6 0 if |η| > c. Here c is a constant of integration which can be determined from the condition (8.11.28), and hence, ∞ c 2 1= v(η) dη = v(η) dη = c3 , (8.11.33) 9 −∞ −c 1
giving c = ( 92 ) 3 . The important point about the solution v(η) is that v (η) does not tend to zero continuously, as η → ∞, as was the case for the linear diffusion problem. The solution v(η) represents a parabola, which intersects the η axis at η = ±c, and its vertex is at (0, 16 c2 ).
408
8 Nonlinear Diffusion–Reaction Phenomena
Thus, the final solution for u(x, t) becomes 1 −1 2 23 t (c t − x2 ) u(x, t) = 6 0
1
if |x| < ct 3 , 1 if |x| > ct 3 .
(8.11.34)
This represents shock wave-like behavior with the propagating wave front at x = 1 2 dx xf = ct 3 . The wave front propagates in the medium with the speed ( dtf ) = 13 ct− 3 which decreases with time t. This means that the wave slows down as t increases. The solution (8.11.34) shows that u is zero ahead of the wave, and its derivative has a jump discontinuity at the front. Recent studies of various nonlinear diffusion models discussed by Murray (1993) and Okubo (1980) for insect and animal dispersal show that grasshoppers exhibit a dispersal behavior similar to that of the above model. Thus, the major conclusion of this analysis is that the solution of the nonlinear diffusion problem is significantly different from the smooth fundamental solution x2 1 exp − (8.11.35) u(x, t) = √ 4κt 4πκt of the linear diffusion problem with a point source at x = 0 and t = 0. Example 8.11.3. Use the Boltzmann transformation η = 12 ( √xt ) to reduce the nonlinear diffusion equation ∂u ∂ ∂u = κ(u) , 0 ≤ x < ∞, t > 0, (8.11.36) ∂t ∂x ∂x with the boundary and initial data u(0, t) = u0 ,
t > 0;
u(x, 0) = u1 ,
0 ≤ x < ∞,
(8.11.37)
to an ordinary differential equation with the independent variable η. Hence, find the solution of the ordinary differential equation. 1 Since η = 12 xt− 2 , we obtain du x −3/2 du 1 and ux = − √ , ut = − t 4 dη 2 t dη and hence, ∂ ∂u ∂ κ du 1 d du √ κ(u) = = κ . ∂x ∂x ∂x 2 t dη 4t dη dη Equation (8.11.36) becomes d du du = κ . −2η dη dη dη
(8.11.38)
8.11 Method of Similarity Solutions of Diffusion Equations
409
The given boundary conditions reduce to u = u0 ,
η = 0,
and
u = u1 ,
as η → ∞.
To find the solution, it is convenient to write (8.11.38) with u ˜= on η = 0 in the form 2η d˜ u d d˜ u − ·κ· = κ . κ dη dη dη
(8.11.39) u u0
and u ˜=1 (8.11.40)
We solve this equation subject to the condition u ˜ = 0, as η → ∞. Integrating equation (8.11.40) twice, we obtain η η 1 2ξ u ˜=1−A exp − dξ dξ, (8.11.41) κ κ 0 0 where A is a constant of integration to be determined, so that η ∞ 1 2ξ exp − dξ dξ = 1, A κ κ 0 0
(8.11.42)
u) is a known in order that u ˜ = uu10 , as η → ∞, is satisfied and where κ = κ(˜ function. The solution (8.11.41) satisfies the condition u ˜ = 1 on η = 0. Considerable attention has been given to the solution of equation (8.11.36) in both semi-infinite (0 < x < ∞) and infinite (−∞ < x < ∞) media for various kinds of concentration-dependent diffusion coefficients. The reader is referred to Chapter 7 of Crank’s book (1975). Example 8.11.4 (Similarity Solution of a More General Nonlinear Diffusion Model). We consider the following nonlinear diffusion equation:
∂ ∂u n ∂u = κ0 u , (8.11.43) ∂t ∂x ∂x where κ0 is a constant. We follow the analysis of Munier et al. (1981) to obtain the similarity solution of (8.11.43). We first introduce the Kirchhoff transformation u un+1 , n + 1 = 0, (8.11.44) tn dt = v= n+1 0 to express (8.11.43) in terms of v as 2 n ∂2v n ∂ v ∂v = κ0 (n + 1)v n+1 2 = κv n+1 2 , ∂t ∂x ∂x
(8.11.45)
n
where κ = κ0 (n + 1) n+1 . We next introduce a one-parameter set of transformations x ˜ = aα x,
t˜ = aβ t,
u ˜ = aγ v,
(8.11.46)
410
8 Nonlinear Diffusion–Reaction Phenomena
to obtain n
n
n
v˜t˜ − κ˜ v n+1 v˜x˜x˜ = aγ−β vt − κa n+1 +γ−2α v n+1 vxx
n = aγ−β vt − κv n+1 vxx ,
(8.11.47)
provided that γn γ−β = + γ − 2α, n+1
that is,
γ = β
n+1 n
2α −1 . β
(8.11.48)
The invariants of the transformations are given by η˜ = x ˜t˜−α/β = xt−α/β = η
and
w ˜ = v˜t˜−γ/β = vt−γ/β = w.
(8.11.49)
Thus, equation (8.11.43) admits the similarity transformation in the form
(8.11.50) v(x, t) = tγ/β w xt−α/β . Substituting (8.11.50) into (8.11.45) gives n n+1 2α α ηw (η) = κw n+1 w (η). −1 w− n β β This equation can be integrated by replacing so that (8.11.51) reduces to n w (η) = −κ0 w− n+1 w +
α β
=
1 n+2 ,
and hence,
1 · ηw , n+1
(8.11.51) γ β
= − n+1 n+2
(8.11.52)
where κ0 = (η + 1)/κ(n + 2). Integrating (8.11.52) once with respect to η gives w (η) = A − κ0 ηw n+1 , 1
(8.11.53)
where A is an integrating constant. The only case in which equation (8.11.53) admits an explicit analytic solution for w corresponds to A = 0. So, in this case, we integrate (8.11.53) to obtain the general solution n nκ0 η 2 n+1 , (8.11.54) w (η) = B − n+1 2 where B is a constant of integration. Hence, the solution for v(x, t) follows from (8.11.50) and (8.11.54) as 2 n n n 1
n x 1 nκ0 − n+2 − n+2 − n+2 n+1 n+1 . (8.11.55) xt = Bt (x, t) = t w − v 2 n+1 t Finally, result (8.11.44) gives u(x, t) in the form 1 1 1 n+1 n 1 nκ0 x2 n n+1 − n+2 u(x, t) = (n + 1)v Bt = (n + 1) − . (8.11.56) 2 n+1 t
8.11 Method of Similarity Solutions of Diffusion Equations
411
Invoking translation and scaling in t, we obtain the final solution
n1 n 1 nκ0 μx2 − n+2 ˜ − , u(x, t) = B(1 + μt) 2 n + 2 (1 + μt)
(8.11.57)
˜ and μ are arbitrary constants. It is noted that the similarity solution in terms where B 1 of η = xt− 2 is possible only when αβ = 12 with an arbitrary n. Example 8.11.5 (Heat Transfer in a Superfluid Helium). At a very low temperature, helium reduces to a liquid phase (called superfluid helium or He-II) with rather unusual properties. One of the most remarkable properties is that heat transport in stationary He-II is described not by the linear Fourier law, but by the nonlinear Gorter– Mellink law 1 ∂T 3 , (8.11.58) q = −κ ∂x where q is the heat flux, κ is a kind of thermal conductivity, and T is the absolute ∂q temperature. Combined with the heat balance equation S( ∂T ∂t ) + ( ∂x ) = 0 and, in suitable units, (8.11.58) gives the nonlinear heat diffusion equation ∂T ∂ = ∂t ∂x
∂T ∂x
13 (8.11.59)
,
where Sκ = 1 and S is the heat capacity per unit volume. This equation is invariant under a one-parameter set of transformations (8.11.4) with α = 1 provided 2γ − 3β = −4. The invariant solutions for the temperature described by the similarity transformations are given by T (x, t) = tγ/β v(η),
η = xt− β , 1
(8.11.60)
where v(η) satisfies an ordinary differential equation d γ dη
dv dη
13 +η
dv − βv = 0. dη
(8.11.61)
This equation is also invariant under the associated transformation group η˜ = μη
and v˜ = μ−2 v,
0 < μ < ∞.
(8.11.62)
The clamped-flux case is of special interest because it has been investigated experimentally in connection with the stability of superconducting magnets cooled with superfluid helium. The initial and boundary conditions for this special problem are given by 3 q Tx = − κ
at x = 0,
and
T (x, t) → 0
as x → ∞, for t > 0, (8.11.63ab)
412
8 Nonlinear Diffusion–Reaction Phenomena
T (x, 0) = 0 for x > 0.
(8.11.64)
We require that γ = 1 and β = 2, so that these conditions become 3 q v (η) = − κ
at η = 0,
and
v(η) → 0,
as η → ∞.
(8.11.65)
In this case, the ordinary differential equation is given by (8.11.61) with γ = 1 and β = 2.
8.12 Nonlinear Reaction–Diffusion Equations The general conservation equation in three space dimensions is ∂u + ∇ · F = f (x, t, u), ∂t
(8.12.1)
where F is a general flux transport due to diffusion or some other processes and f (x, t, u) is the source or reaction term. For the case of general diffusion processes, we can take F = −κ∇u, so that equation (8.12.1) becomes ∂u = ∇ · (κ∇u) + f (x, t, u), ∂t
(8.12.2)
where κ = κ(x, u) is a function of x and u. For the case of several chemicals or interacting species, the vector ui (x, t), i = 1, 2, . . . , n, represents concentrations or densities each diffusing with its own diffusion coefficient κi and interacting according to the vector source term f (x, u, t). Then, equation (8.12.2) becomes ∂u = ∇ · (κ∇u) + f (x, t, u), ∂t
(8.12.3)
where κ represents a diffusivity matrix with no cross-diffusion among the species, so that κ simply becomes a diagonal matrix. In fact, ∇u is a tensor, so that ∇ · κ∇u becomes a vector. Equation (8.12.3) represents a system of nonlinear (or interacting population–diffusion) equations. Turin (1952) used it as an important model for the chemical basis of morphogenesis. Such systems have been widely investigated in the 1970s. One simple model equation as a special case of (8.12.3) is ∂u = ∇ · (κ∇u) + f (u), ∂t
(8.12.4)
where κ is a diagonal matrix. In particular, when κ is independent of x and u, equation (8.12.4) reduces to the form ∂u = κ∇2 u + f (u). (8.12.5) ∂t
8.12 Nonlinear Reaction–Diffusion Equations
413
Physically, the components of u represent the concentrations of certain species, which are reacting with each other at each point and also diffusing through the medium. Further, if f (u) = 0, (8.12.5) becomes a system of diffusion equations ∂u = κ∇2 u. ∂t
(8.12.6)
This is regarded as a mathematical model of the classical kinetics of systems of reaction. For more information, the reader is referred to Kopell and Howard (1973), who studied plane wave solutions of reaction–diffusion equations. In the one-dimensional case with f as a function of u only, we can write (8.12.2) as
∂u ∂ ∂u = κ(u) + f (u), (8.12.7) ∂t ∂x ∂x where usually κ(u) = kur , and k and r are positive constants. This equation with f ≡ 0 has been studied much more widely than that of the nonzero f . However, it is of interest to examine equation (8.12.7) with functions f (u) with two simple zeros, one at u = 0 and the other at u = 1. We write another version of (8.12.7) with f (u) = aum (1 − un ), where a, m, and n are positive constants in the form
∂u ∂u ∂ =k ur + aum 1 − un . (8.12.8) ∂t ∂x ∂x Pattle (1959) solved this equation without the reaction term (a = 0). On the other hand, Newman (1980) obtained a number of traveling wave solutions u(ξ), ξ = x − ct, for various values of r and m with n = 1. His study shows that, when r > 0, u(ξ) = 0 at a prescribed positive value of ξ. Furthermore, when r = 0 and m = 2, the solution represents an ordinary logistic distribution. According to Newman, this model equation can be applied to problems with growth phenomena, and also to population genetics and combustion processes. Montroll and West (1973) obtained a number of solutions for various problems involving diffusion and growth. a x and t∗ = at to It is convenient to use nondimensional variables x∗ = k rewrite (8.12.6) in nondimensional form, dropping the asterisks,
∂u ∂u ∂ = ur + um 1 − un . (8.12.9) ∂t ∂x ∂x With u(x, t) = u(ξ), ξ = x − ct, we obtain the traveling wave solution of the equation
du du d ur +c + um 1 − un = 0. (8.12.10) dξ dξ dξ Several authors including Kaliappan (1984) and Murray (1993) obtained an exact traveling wave solution of (8.12.10) for the case where r = 0 and m = 1 with the boundary conditions u(−∞) = 1
and u(+∞) = 0.
(8.12.11)
414
8 Nonlinear Diffusion–Reaction Phenomena
Making reference to Murray’s (1993) detailed analysis with r = 0 and m = 1, we seek an exact solution of (8.12.10) in the form u(ξ) =
1 , (1 + aebξ )s
(8.12.12)
where a, b, and s are positive constants to be determined. This solution automatically satisfies the boundary conditions (8.12.11) at ξ = ±∞. We substitute (8.12.12) in (8.12.10) to obtain the following results: s=
2 , n
u(0) = u0 ,
n , b= 2(n + 2) −n 2
a = u0
c=
n+4 2(n + 2)
,
− 1.
(8.12.13) (8.12.14)
The second derivative of solution (8.12.12) gives the point of inflection ξi =
n 1 log , b 2a
ui =
− n2 n 1+ . 2
(8.12.15)
The above analysis provides a more general traveling wave solution of the power law logistic equation with diffusion. In particular, when r = m = 0 and n = 1, (8.12.10) reduces √ to the Fisher equation which was solved earlier. If we choose u0 = 1 , then a = 2 − 1, and hence, all quantities b, c, ξi , and ui can be calculated, and 2 solution (8.12.12) is known exactly. If we set n = 2 in (8.12.13)–(8.12.15), then s = 1, b = √12 , c = √32 , a = 1, ξi = 0, and ui = 12 . In this case, the solution (8.12.12) reduces to that of the ordinary logistic equation with diffusion. When r = 0 and m = n + 1, equation (8.12.9) becomes
∂2u ∂u = + un+1 1 − un . 2 ∂t ∂x
(8.12.16)
Following Murray (1993), we obtain the traveling wave solution u(ξ) with ξ = x − ct of equation (8.12.16) in the form u(ξ) =
1 , (1 + aebξ )s
(8.12.17)
n 1 , and c = √n+1 . where s = n1 , b = √n+1 When n = 1, equation (8.12.16) is further simplified, and then, s = 1, b = c = √12 , and the resulting solution obtained from (8.12.17) is in agreement with Newman’s (1980) solution in population genetics. This solution also describes the situation in which the reaction term is modified by the coalition growth instead of the logistic growth. Another interesting and useful exact solution was obtained by Murray (1993) for the case r = m = n = 1. This case describes density-dependent diffusion with logistic population growth, which is represented by the equation
8.12 Nonlinear Reaction–Diffusion Equations
∂u ∂ ∂u = u + u(1 − u). ∂t ∂x ∂x
415
(8.12.18)
We seek a traveling wave solution u(x, t) = u(ξ), ξ = x − ct of (8.12.18) with u(−∞) = 1 and u(+∞) = 0. Thus, equation (8.12.18) reduces to (uu ) + cu + u(1 − u) = 0,
(8.12.19)
where the prime denotes the derivative with respect to ξ. Thus, the associated phase plane system is given by du = v, dξ dv = − cv + v2 + u(1 − u) . u dξ
(8.12.20a) (8.12.20b)
In (8.12.20b), u = 0 is a singularity which can be removed by defining a new variable d d = u dξ , so that equations (8.12.20a)–(8.12.20b) become ζ by dζ du = uv and dζ dv = − cv + v 2 + u(1 − u) . dζ
(8.12.21a) (8.12.21b)
Clearly, the second equation is not singular. The singular points in the (u, v)-phase plane are (u, v) = (0, 0), (1, 0), (0, −c). A phase plane analysis shows that (1, 0) and (0, −c) are saddle points, whereas (0, 0) is a stable nonlinear node because of the nonlinear term uv in (8.12.21a). Murray (1993) described the phase trajectories of (8.12.21a)–(8.12.21b) for different values of c with graphical representations. Basically, the model equation (8.12.18) indicates that the population disperses to regions of lower density more rapidly as the population increases. The traveling wave solution u(ξ)with boundary conditions u(−∞) = 1, u(+∞) = 0, and c = √12 is given by ξ − ξc H(ξc − ξ), (8.12.22) u(ξ) = 1 − exp √ 2 where ξc is the wave front and H(x) is the Heaviside unit step function. Several authors including Aronson (1980), Crank (1975), Fife (1979), Ghez (1988), Gurney and Nisbet (1975), Gurtin and MacCamy (1977), Newman (1980), and Shigesada (1980) have studied various problems of density-dependent diffusion and their extensions. The above analysis, showing the existence of the traveling wave solutions, can be generalized to more general, nonlinear, diffusion phenomena in which the diffusion coefficient is ur , for r = 1, or, even more general, κ(u) in (8.12.4) with certain conditions imposed on κ(u). In general, nonlinear reaction–diffusion interactions radically change the mathematical and physical features of the solutions of many biological and chemical problems.
416
8 Nonlinear Diffusion–Reaction Phenomena
Further, we write out the diffusion term in (8.12.4) in full to obtain ut = f (u) + κ (u)u2x + κ(u)uxx .
(8.12.23)
This shows that the nonlinear diffusion term can be thought of as an equivalent convection with velocity h (u) = −κ (u)ux . If the convection arises as a natural extension of a conservation law, we can derive an equation with the convection velocity h (u) in the form ∂ 2u ∂u + h (u)ux = f (u) + κ(u) 2 . (8.12.24) ∂t ∂x Such equations arise in a wide variety of problems in physical, chemical, and biological sciences. It is important to point out that the nonlinear convection in reaction– diffusion equations can have a significant effect on the solutions. This is to be expected since there is another major transport process, namely, linear or nonlinear convection. This process may or may not dominate over diffusion effects. However, if the effect of diffusion is negligible compared with that of convection, convection obviously tends to make parts of the wave form grow steeper and steeper, and solutions develop discontinuities in the form of shock waves. Indeed, the development of shock waves and the breaking of a wave form are typical nonlinear convection phenomena.
8.13 Brief Summary of Recent Work To understand various physical phenomena described by the interaction of diffusion, convection, or relaxation, Lighthill (1956) gave a fully self-contained account of the basic physical features of shock wave development, propagation, and decay, the internal structure of shocks, and their confluence with the derivation of a nonlinear coupled system intermediate between the Navier–Stokes equations and the Burgers equation. He then discussed approximate solutions of several physical problems by using graphical construction of characteristics and the method of steepest descent. He also showed that the Burgers equation is the appropriate one for the investigation of weak planar-wave phenomena. Subsequently, many authors including Blackstock (1964), Walsh (1969), Benton and Platzman (1972), Crighton and Scott (1979), Parker (1980), Rodin (1970), Larson (1978), and Lardner (1986) studied initial boundary-value problems for the Burgers equation and investigated the physical significance of their solutions. It is pertinent to mention that Benton and Platzman (1972) have also compiled an exhaustive list of the possible solutions of the Burgers equation and have illustrated the physically interesting ones by isochronal graphs. At the same time, considerable attention has been given to the determination of self-similar solutions as intermediate asymptotics for nonlinear diffusion equations (see Barenblatt and Zel’dovich 1972; Zel’dovich and Raizer 1966, 1968; Barenblatt 1979, and Newman 1983). Many generalizations of the Burgers equation have been made by several authors including Case and Chiu (1969), Murray (1970a, 1970b, 1973), Penel and Brauner (1974), and Crighton (1979). Kriess
8.14 Exercises
417
and Lorenz (1989) discussed the existence and uniqueness of solutions of boundaryvalue problems associated with the Burgers equation. Recent developments in the reaction–diffusion equation in general and the Fisher equation in particular are available in papers by Johnson (1970), McKean (1975), Barenblatt (1979), Manoranjan and Mitchell (1983), Hagstrom and Keller (1986), Gazdag and Canosa (1974), Murray (1993), Smoller (1994), Britton (1986), Grindrod (1991), Dunbar (1983), Logan and Dunbar (1992), and Fife (1979). The Fisher equation also represents a nonlinear model equation for the evolution of a neutron population in a nuclear reactor. Canosa (1969, 1973) showed that the Fisher equation describes a balance between linear diffusion and nonlinear local multiplication, and it admits shock-type solutions. Gazdag and Canosa (1974) used the pseudospectral method accurately to discretize spatial derivatives for numerically solving the Fisher equation. Tang and Webber (1991) presented an interesting and precise numerical study of the Fisher equation by a Petrov–Galerkin finite element method. Their analysis shows that any local initial disturbance can propagate with a constant limiting speed as time tends to infinity. Both the limiting wave fronts and limiting speed are determined by the system itself and are independent of the initial states. Compared with other numerical methods, Tang and Webber’s numerical study is more satisfactory with regard to its accuracy and stability. Larson (1978) studied a more general Fisher equation and determined lower and upper bounds of its solutions. Ablowitz and Zeppetella (1979) have investigated the exact solution and Painlevé transcendents. They have also shown that equation (8.9.2) can be transformed into a simpler nonlinear equation which admits solutions in terms of the Weierstrass ℘ function. For the generalized Fisher equation with f (u) in place of u(1 − u), Kametaka (1976) proved the existence theorem for the traveling wave solution when certain conditions on f (u) hold, and he investigated the stability of traveling wave and transient solutions. Finally, we refer to the work of Abdelkader (1982), who extended the results of Ablowitz and Zeppetella (1979) for the generalized Fisher equation of the form u + cu + u − un = 0,
1 < n < ∞.
(8.13.1)
For c = (n + 3)/ (2n + 2), 2 < c < ∞, the solution of (8.13.1) can be expressed in terms of hyperelliptic integrals. In some special cases, the solution was obtained by Rosenau (1982) in terms of elliptic integrals for a related equation describing nonlinear thermal wave phenomena in a reacting medium.
8.14 Exercises 1. Determine a set of stretching transformations (8.11.4) under which the following equations are invariant: (a) utt = c2 uxx , (c) ut + uux = 0, (e) uut = uxx (x > 0, t > 0),
(b) uut + u2x = 0, (d) ut = κ(urr + 2r ur ), (f) c2 utt + uxxxx = 0 (c2 =
EI m ).
418
8 Nonlinear Diffusion–Reaction Phenomena
2. Use ξ = x−ct to reduce the reaction–diffusion equation (see Jones and Sleeman 1983) ut − uxx = f (u) into the form
du d2 u + f (u) = 0. +c dξ 2 dξ
Solve this ordinary differential equation with the reaction term u on D1 : 0 ≤ u ≤ 12 , f (u) = 1 − u on D2 : 12 ≤ u ≤ 1 and the boundary conditions u(−∞) = 1 and
u(+∞) = 0.
Discuss the solution for cases (i) c = 2 and (ii) c = 2. 3. Show that the similarity transformation of the Boltzmann nonlinear equation (8.11.20) with boundary conditions (uux )x=0 = −a = const., u(x, t) → 0 u(x, 0) = 0 for x > 0, is
1
u(x, t) = t 3 v(η),
as x → ∞, t > 0,
η = xt− 3 , 2
where v(η) satisfies the nonlinear ordinary differential equation 3(vv ) + 2ηv − v = 0. 4. Show that the Boltzmann equation (8.11.20) is invariant under the one-parameter ˜ set of transformations x ˜ = x exp( ατ ˜ = u exp(ατ ), 0 < τ < ∞. 2 ), t = t + τ , u The boundary condition (Barenblatt and Zel’dovich 1972) u(0, t) = u0 exp(t),
−∞ < t < ∞,
is also invariant under the above set of transformations when α = 1. Show that the most general invariant solution has the form αt u(x, t) = exp(αt)f x exp − . 2 When α = 1, u(x, t) = exp(t)v{x exp(− 2t )}. If it is a solution of (8.11.20), then its image under the above set of transformations must be a solution, that is, ατ t+r exp − eατ u(x, t) = v(η) exp(t + r), η = x exp 2 2 is also a solution.
8.14 Exercises
419
5. Show that the similarity solution of the Burgers equation (8.2.6) with v = 1 is u(x, t) = (2t + a)−2 f (η),
η = (2t + a)− 2 x, 1
where f (η) satisfies the nonlinear ordinary differential equation f − (f − η)f + f = 0. 6. Show that the similarity solution of the Burgers equation (8.2.6) with ν = 1 is given by 1 1 u(x, t) = √ f (η) − 1, η = (x + t)t− 2 , t where f satisfies the equation
f +f
η −f 2
1 + f = 0. 2
7. The Prandtl–Blasius boundary-layer problem for a flat plate can be described by the equations uux + vuy = νuyy , ux + vy = 0, with the boundary conditions (u, v) = (0, 0) at y = 0, x > 0, u = U as y → ∞, x > 0, u = U at x = 0, y > 0, where U is the incident velocity. Show that these equations are invariant under the set of transformations x ˜ = aα x,
u ˜ = aβ u,
y˜ = ay,
and v˜ = a−1 v,
where α − β = 2. Show also that the boundary conditions are all invariant under the above transformations when α = 2 and β = 0. Show that the first two boundary conditions are also invariant even if β = 0 and the last two boundary conditions become u ˜ = aβ U . 8. In terms of the stream function ψ defined by (u, v) = (ψy , −ψx ) with ν = 1 and U = 1, the Prandtl–Blasius equation in Exercise 7 is given by ψyyy = ψy ψxy − ψx ψyy . Show that the set of similarity transformations of this equation is 1
ψ(x, y) = x 2 f (η),
η = yx− 2 , 1
420
8 Nonlinear Diffusion–Reaction Phenomena
where f (η) satisfies the Blasius equation 2f + f f = 0 with boundary conditions f (0) = 0,
f (0) = 0,
and f (∞) = 1.
9. The Sparrow et al. (1970) problem of a flat plate with uniform suction or in jection is obtained by changing the variables (ξ, η) = (vω Ux , y Ux ) in the Prandtl–Blasius equation in Exercise 8, where vω is the constant suction velocity. Show that ξ and η are invariant under the set of transformations introduced in Exercise 7. Show that the similarity transformation of this problem with ν = 1 1 and U = 1 is given by ψ = x 2 f (ξ, η), where f satisfies the equation 2fηηη + f fηη = ξ(fη fξη − fξ fηη ) with boundary conditions fη (η = 0) = 0,
fη (η → ∞) = 1,
and
f + 2ξ = −ξfξ
at η = 0.
10. In Exercise 1(e), show that v(η) = t
2α β −1
u(x, t),
η = xβ t−α
are invariant under (8.11.4). Using the data u(x, 0) = 0, x > 0, u(x, t) → 0, as x → ∞, t > 0, and ux (0, t) = −q, t > 0, where q is a constant, show that β = 1 and α = 13 . Hence, show that v(η) satisfies the problem 3v − v(v − ηv ) = 0, 0 < η < ∞, v (0) = −q, and v(η) → 0
as η → ∞.
11. In Exercise 1(f), show that the quantities v(η) = t− β u γ
and
η = xt−α/β
are invariant under (8.11.4). Hence prove that v(η) satisfies the ordinary differential equation 2 α γ α α 2γ γ 2 2 v +c η v + 1+ − ηv + −1 v = 0. β β β β β β Derive the equation for 12. Show that equations
α β
=
1 2
= − βγ .
(a) ut + uux + μuxxx = 0
and (b) ut + up ux + μuxxx = 0
are invariant under the transformations x ˜ = aα x,
t˜ = aβ t,
and u ˜ = au,
provided that α = − 12 and β = − 32 for equation (a), and α = − p2 and β = − 3p 2 for equation (b).
8.14 Exercises
421
13. Consider the nonlinear reaction–diffusion model ut − κuxx = F (u), 0 < x < a, t > 0, ux (0, t) = 0 = ux (a, t), t > 0, u(x, 0) = f (x),
0 < x < a,
where the reaction term F is continuously differentiable and supu∈R |F (u)| = A < ∞. Show that the quantity a
u2x (x, t) dx
Q(t) = 0
associated with this model satisfies the inequality Q(t) ≤ Q(0) exp(μt),
μ = 2 A − π2 .
If A < π 2 , show that Q(t) → 0 as t → ∞. 14. Consider the coupled model equations 2u u t = κ ux − cx , c x
ct + ku = 0,
where u = u(x, t) is the population of an organism, c = c(x, t) is the concentration of a nutrient to which the organisms are attracted, and k is a positive constant. Obtain the traveling wave solutions in the form u = u(x − U t)
and
c = c(x − U t),
where U (±∞) = 0,
c(−∞) = 0,
and c(+∞) = 1.
Draw the graph of the solutions and give a biological interpretation of the solutions. 15. If av = (1 − b)(1 − u), a > 0, 0 < b < 1, show that the reaction–diffusion system ut − uxx = u(1 − u − av), vt − vxx + buv = 0 reduces to the Fisher equation. 16. Consider a simple model of a detonation process governed by the system ut + uux + λx = κuxx ,
λt = r(u, λ),
where u is a scaled temperature and λ is the mass fraction of the product species P in a reversible chemical reaction R ↔ P with the reaction rate r = 1 − λ − λ exp(−u−1 ), and where κ is a positive constant. Show that there exist positive traveling wave solutions if the wave speed c exceeds the value of u at +∞ (Logan and Dunbar 1992).
422
8 Nonlinear Diffusion–Reaction Phenomena
17. Show that the traveling wave solution exists for the wave velocity c > 2 for the reaction–convection system (Logan and Shores 1993a, 1993b): vt + v = 1 − v exp(u−1 ).
ut + uux + vx = (2 − u)(u − 1),
18. (a) Show that the transformation v(x, t) = exp(−αt)u(x, t) reduces the equation vt + αv = κuxx , into the form ut = κuxx . (b) Find a change of variable that reduces the diffusion equation ut = κuxx , into the form vt = vxx . 19. Use the transformations a(τ ) dτ
and v(ξ, t) = u ξ +
t
a(τ ) dτ, t 0
to show that the equation ut + a(t)ux = κuxx , can be transformed into the form vt = κvxx . 20. Using the conservation of density (8.2.1) with q = ρu and the conservation of momentum, qt + (uq)x = 0, derive the inviscid Burgers equation 1 2 u ut + uux = 0, or equivalently, ut + = 0. 2 x Hence, or otherwise, show that the solution ux at any time t is (ux )(t) = where u0 = (ux )t=t0 .
u0 , 1 + (t − t0 )u0
8.14 Exercises
423
21. The Fokker–Planck equation in non-equilibrium statistical mechanics for the probability distribution function u(x, t) is a modified diffusion equation ut = uxx + (xu)x ,
x ∈ R, t > 0.
By separation of variables u = X(x)T (t), show that X (x) + xX + (1 + n)X = 0 and T˙ + nT = 0. where n is the separation constant. Assuming X(x) = f (x) exp(− 12 x2 ) and rescaling the independent variables √ (x = 2ξ), show that f (ξ) = Hn (ξ) satisfies the equation d2 f df − 2ξ + 2nf = 0. dξ 2 dξ Show that the solution u(x, t) of the Fokker–Planck equation is 1 2 x an exp −nt − x Hn √ . u(x, t) = 2 2 n=0 ∞
Find the limiting form of the solution u(x, t) as t → ∞.
9 Solitons and the Inverse Scattering Transform
True Laws of Nature cannot be linear. Albert Einstein . . . the great primary waves of translation cross each other without change of any kind in the same manner as the small oscillations produced on the surface of a pool by a falling stone. Scott Russell
9.1 Introduction Dispersion and nonlinearity play a fundamental role in wave motions in nature. The nonlinear shallow water equations that neglect dispersion altogether lead to breaking phenomena of the typical hyperbolic kind with the development of a vertical profile. In particular, the linear dispersive term in the Korteweg–de Vries equation prevents this from ever happening in its solution. In general, breaking can be prevented by including dispersive effects in the shallow water theory. The nonlinear theory provides some insight into the question of how nonlinearity affects dispersive wave motions. Another interesting feature is the instability and subsequent modulation of an initially uniform wave profile. To understand these features, this chapter is devoted to the Boussinesq and Korteweg–de Vries (KdV) equations and solitons with emphasis on the methods and solutions of these equations that originated from water waves. Special attention is given to the inverse scattering transform, conservation laws and nonlinear transformations, Bäcklund transformations and the nonlinear superposition principle, the Lax formulation and its KdV hierarchy, the ZS (Zakharov and Shabat) scheme, and the AKNS (Ablowitz, Kaup, Newell, and Segur) method. Finally, we consider a new class of solitary waves with compact support which are called compactons governed by a two-parameter family of strongly dispersive nonlinear equations, K(m, n). Included are the existence of peakon (singular) solutions of a new strongly nonlinear L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_9,
426
9 Solitons and the Inverse Scattering Transform
model in shallow water described by the Camassa and Holm equation, and the Harry Dym equation which arises as a generalization of the class of isospectral flows of the Schrödinger operator. As an example of the application of compactons, the solution of nonlinear vibration of an anharmonic mass–spring system is presented. This is followed by a discussion on the existence of new nonlinear intrinsic localized modes in anharmonic crystals. Based on the rotating-wave approximation, Sievers and Takeno (1988) discovered the s-localized modes, while Page (1990) introduced the p-localized modes in a one-dimensional lattice model. These nonlinear localized modes may be treated as compactons.
9.2 The History of the Solitons and Soliton Interactions Historically, John Scott Russell first experimentally observed the solitary wave, a long water wave without change in shape, on the Edinburgh–Glasgow Canal in 1834. He called it the “great wave of translation” and then reported his observations at the British Association in his 1844 paper “Report on Waves.” Thus, the solitary wave represents, not a periodic wave, but the propagation of a single isolated symmetrical hump of unchanged form. His discovery of this remarkable phenomenon inspired him further to conduct a series of extensive laboratory experiments on the generation and propagation of such waves. Based on his experimental findings, Russell discovered, empirically, one of the most important relations between the speed U of a solitary wave and its maximum amplitude a above the free surface of liquid of finite depth h in the form (9.2.1) U 2 = g(h + a), where g is the acceleration due to gravity. His experiments stimulated great interest in the subject of water waves and his findings received a strong criticism from Airy (1845) and Stokes (1847). In spite of his remarkable work on the existence of periodic wavetrains representing a typical feature of nonlinear dispersive wave systems, Stokes’ conclusion on the existence of the solitary wave was erroneous. However, Stokes (1847) proposed that the free surface elevation of the plane wavetrains on deep water can be expanded in powers of the wave amplitude. His original result for the dispersion relation in deep water is (9.2.2) ω 2 = gk 1 + a2 k 2 + · · · . Stokes’ theory predicts that periodic wavetrains are possible in nonlinear dispersive systems, and the dispersion relation involves the wave amplitude, which produces significant qualitative changes in water wave phenomena. In spite of the introduction of the Stokes expansion, definite analytical proof of the existence of a solution representing permanent water waves has been a formidable task. During the first quarter of the twentieth century, there has been a serious controversy on this issue, and, in fact, real doubt was raised about the convergence of the Stokes expansion. The problem was eventually resolved by Levi-Civita (1925), who proved formally
9.2 The History of the Solitons and Soliton Interactions
427
Fig. 9.1 A solitary wave.
that the Stokes expansion for deep water waves converges provided the wave amplitude is very small compared with the wavelength. Almost simultaneously, Struik (1926) extended the proof of convergence to small-amplitude waves on water of finite depth. Subsequently, Krasovskii (1960, 1961) established the existence of permanent periodic waves for all values of amplitude less than the maximum value at which the waves assume a sharp-crested form. Finally, rigorous proofs of the existence of water waves of greatest height have been given by Toland (1978) and Keady and Norbury (1978). Despite these serious attempts to prove the existence of finiteamplitude water waves of permanent form, the independent question of their stability remained unattended until the 1960s except for an isolated study by Korteweg and de Vries in (1895) on long surface waves in water of finite depth. But one of the most remarkable discoveries made in the 1960s was that the periodic Stokes waves on sufficiently deep water are definitely unstable! This result seems revolutionary in view of the sustained attempts to prove the existence of Stokes waves of finite amplitude and permanent form. Russell’s description of solitary waves contradicted the theories of water waves due to Airy and Stokes; they raised questions on the existence of Russell’s solitary waves and conjectured that such waves cannot propagate in a liquid medium without change of form. It was not until the 1870s that Russell’s prediction was finally and independently confirmed by both Boussinesq (1871a, 1871b, 1872, 1877) and Rayleigh (1876). From the equations of motion for an inviscid incompressible liquid, they derived formula (9.2.1). In fact, they also showed that the solitary wave profile (see Figure 9.1) z = η(x, t) is given by (9.2.3) η(x, t) = a sech2 β(x − U t) , where β 2 = 3a div{4h2 (h + a)} for any a > 0. Although these authors found the sech2 solution, which is valid only if a h, they did not write any equation for η that admits (9.2.3) as a solution. However, Boussinesq did a lot more and discovered several new ideas, including a nonlinear evolution equation for such long water waves in the form 3 η2 1 2 2 (9.2.4) + h ηxxx , ηtt = c ηxx + 2 h xx 3
428
9 Solitons and the Inverse Scattering Transform
√ where c = gh is the speed of the shallow water waves. This is known as the Boussinesq (bidirectional) equation, which admits the solution 1/2 η(x, t) = a sech2 3a/h3 (x ± U t) .
(9.2.5)
This represents solitary waves traveling in both the positive and negative x-directions. More than 60 years later, in 1895, two Dutchmen, D.J. Korteweg and G. de Vries, formulated a mathematical model equation to provide an explanation of the phenomenon observed by Scott Russell. They derived the now-famous equation for the propagation of waves in one direction on the surface water of density ρ in the form 3 c 1 (9.2.6) ε + η ηX + σηXXX , ηt = h 2 2 √ where X is a coordinate chosen to be moving (almost) with the wave, c = gh, ε is a small parameter, and 2 h T 1 − ∼ h3 , σ=h (9.2.7) 3 gρ 3 when the surface tension T ( 13 gρh2 ) is negligible. Equation (9.2.6) is known as the Korteweg–de Vries (KdV) equation. This is one of the simplest and most useful nonlinear model equations for solitary waves, and it represents the longtime evolution of wave phenomena in which the steepening effect of the nonlinear term is counterbalanced by linear dispersion. It is convenient to introduce the change of variables η = η(X ∗ , t) and X ∗ = X + (ε/h)ct which, dropping the asterisks, allows us to rewrite equation (9.2.6) in the form c 3 1 (9.2.8) ηηX + σηXXX . ηt = h 2 2 Modern developments in the theory and applications of the KdV solitary waves began with the seminal work published as a Los Alamos Scientific Laboratory Report in 1955 by Fermi, Pasta, and Ulam on a numerical model of a discrete nonlinear mass–spring system. In 1914, Debye suggested that the finite thermal conductivity of an anharmonic lattice is due to the nonlinear forces in the springs. This suggestion led Fermi, Pasta, and Ulam to believe that a smooth initial state would eventually relax to an equipartition of energy among all modes because of nonlinearity. But their study led to the striking conclusion that there is no equipartition of energy among the modes. Although all the energy was initially in the lowest modes, after flowing back and forth among various low-order modes, it eventually returns to the lowest mode, and the end state is a series of recurring states. This remarkable fact has become known as the Fermi–Pasta–Ulam (FPU) recurrence phenomenon. Cercignani (1977) and later on Palais (1997) described the FPU experiment and its relationship to the KdV equation in some detail. This curious result of the FPU experiment inspired Martin Kruskal and Norman Zabusky to formulate a continuum model for the nonlinear mass–spring system to
9.2 The History of the Solitons and Soliton Interactions
429
Fig. 9.2 Development of solitary waves: (a) initial profile at t = 0, (b) profile at t = π −1 , and (c) wave profile at t = (3.6)π −1 . From Zabusky and Kruskal (1965).
understand why recurrence occurred. In fact, they considered the initial-value problem for the KdV equation, ut + uux + δuxxx = 0,
(9.2.9)
where δ = ( h )2 , is a typical horizontal length scale, with the initial condition u(x, 0) = cos πx,
0 ≤ x ≤ 2,
(9.2.10)
and the periodic boundary conditions√with period 2, so that u(x, t) = u(x + 2, t) for all t. Their numerical study with δ = 0.022 produced a lot of new interesting results which are shown in Figure 9.2. They observed that, initially, the wave steepened in regions where it had a negative slope, a consequence of the dominant effects of nonlinearity over the dispersive term, δuxxx . As the wave steepens, the dispersive effect then becomes significant and balances the nonlinearity. At later times, the solution develops a series of eight welldefined waves, each like sech2 functions with the taller (faster) waves ever catching up and overtaking the shorter (slower) waves. These waves undergo nonlinear interaction according to the KdV equation and then emerge from the interaction without change of form and amplitude, but with only a small change in their phases. So, the most remarkable feature is that these waves retain their identities after the nonlinear interaction. Another surprising fact is that the initial profile reappears, very similarly to the FPU recurrence phenomenon. In view of their preservation of shape and the resemblance to the particle-like character of these waves, Kruskal and Zabusky called these solitary waves, solitons, like photon, proton, electron, and other terms for elementary particles. Historically, the famous 1965 paper of Zabusky and Kruskal marked the birth of the new concept of the soliton, a name intended to signify particle-like quantities. Subsequently, Zabusky (1967) confirmed, numerically, the actual physical interaction of two solitons, and Lax (1968) gave a rigorous analytical proof that the identities of two distinct solitons are preserved through the nonlinear interaction governed
430
9 Solitons and the Inverse Scattering Transform
by the KdV equation. Physically, when two solitons of different amplitudes (and hence, of different speeds) are placed far apart on the real line, the taller (faster) wave to the left of the shorter (slower) wave, the taller one eventually catches up to the shorter one and then overtakes it. When this happens, they undergo a nonlinear interaction according to the KdV equation and emerge from the interaction completely preserved in form and speed with only a phase shift. Thus, these two remarkable features, (i) steady progressive pulse-like solutions and (ii) the preservation of their shapes and speeds, confirmed the particle-like property of the waves and, hence, the definition of the soliton. Subsequently, Gardner et al. (1967, 1974) and Hirota (1971, 1973a, 1973b) constructed analytical solutions of the KdV equation that provide the description of the interaction among N solitons for any positive integral N . After the discovery of the complete integrability of the KdV equation in 1967, the theory of the KdV equation and its relationship to the Euler equations of motion as an approximate model derived from the theory of asymptotic expansions became of major interest. From a physical point of view, the KdV equation is not only a standard nonlinear model for long water waves in a dispersive medium, it also arises as an approximate model in numerous other fields, including ion-acoustic plasma waves, magnetohydrodynamic waves, and anharmonic lattice vibrations. Experimental confirmation of solitons and their interactions has been provided successfully by Zabusky and Galvin (1971), Hammack and Segur (1974), and Weidman and Maxworthy (1978). Thus, these discoveries have led, in turn, to extensive theoretical, experimental, and computational studies over the last 30 years. Many nonlinear model equations have now been found that possess similar properties, and diverse branches of pure and applied mathematics have been required to explain many of the novel features that have appeared. Finally, it seems pertinent to mention the following. It is not easy to give a precise definition of a soliton. However, the term can readily be associated with any solution of a nonlinear partial differential equation (or system) that (i) represents a wave of permanent form, (ii) is localized, so that it decays or approaches a constant value at infinity, and (iii) can undergo a strong interaction with other solitons and retain its identity. In the context of the KdV equation and other similar equations, the singlesoliton solution is usually referred to as the solitary wave, but when more than one of them appear in a solution, they are called solitons. In other words, a soliton is a solitary wave when it is infinitely separated from any other soliton. Also, for equations other than the KdV equation, the solitary-wave solution may not be a sech2 function, but a sech or tan−1 (eαx ) profile. In fact, some nonlinear equations have solitarywave solutions but not solitons, whereas others (like the KdV equation) have solitary waves that are solitons. Indeed, the soliton has formed a new paradigm in mathematical physics. In recent years, the concept of the soliton has been used in a very broad sense. For example, the nonlinear Schrödinger equation (NLS), describing waves in plasmas, in superconductors, and in nonlinear optics, etc., yields the envelope soliton solution which has the form of a sech profile modulating a monochromatic carrier wave. Unlike the KdV solitons, an NLS soliton does not depend on the amplitude. Physically, the former is called a low-frequency soliton, whereas the latter is called a high-frequency soliton (Makhankov 1978). Another important model equation, the
9.3 The Boussinesq and Korteweg–de Vries Equations
431
Fig. 9.3 A shallow water wave model.
sine-Gordon (SG) equation, is used to describe nonlinear wave motions in the unified theory of elementary particles, magnetic flows, and dislocations in crystals. This equation admits SG soliton solutions that are called either kinks or antikinks, whose velocities do not depend on the wave amplitude.
9.3 The Boussinesq and Korteweg–de Vries Equations We consider an inviscid liquid of constant mean depth h and constant density ρ without surface tension. We assume that the (x, y)-plane is the undisturbed free surface with the z-axis positive upward. The free surface elevation above the undisturbed mean depth h is given by z = η(x, y, t), so that the free surface is at z = H = h + η and z = 0 is the horizontal rigid bottom (see Figure 9.3). It has already been recognized that the parameters ε = a/h and κ = ak, where a is the surface wave amplitude and k is the wavenumber, must both be small for the linearized theory of surface waves to be valid. To develop the nonlinear shallow water theory, it is convenient to introduce the following nondimensional flow variables based on a different length scale h (which could be the fluid depth): ∗ ∗ 1 z ct η ∗ ∗ , η∗ = , z = , t = x , y = (x, y), l h l a (9.3.1) h φ∗ = φ, alc √ where l is a typical horizontal length scale, and c = gh is the typical horizontal velocity (or shallow water wave speed). We next introduce two fundamental parameters to characterize the nonlinear shallow water waves: h2 a and δ = 2 , ε= (9.3.2ab) h l √ where ε is called the amplitude parameter and δ is called the long wavelength or shallowness parameter.
432
9 Solitons and the Inverse Scattering Transform
In terms of the preceding nondimensional variables and the parameters, the basic equations for water waves (Debnath 1994, equations (1.6.5)–(1.6.8)) can be written in the nondimensional form, dropping the asterisks: δ(φxx + φyy ) + φzz ε ∂φ ε 2 + φx + φ2y + φ2z + η ∂t 2 2δ δ ηt + ε(φx ηx + φy ηy ) − φz φz
= 0,
(9.3.3)
= 0 on z = 1 + εη,
(9.3.4)
= 0 on z = 1 + εη, = 0 on z = 0.
(9.3.5) (9.3.6)
It is noted that the parameter κ = ak does not enter explicitly in equations (9.3.3)–(9.3.6), but an equivalent √ parameter γ = a/l is associated with ε and δ through γ = (a/h) · (h/l) = ε δ. If ε is small, the terms involving ε in (9.3.4), (9.3.5) can be neglected to recover the linearized free surface conditions. However, the assumption that δ is small might be interpreted as the characteristic feature of the shallow water theory. So, we expand φ in terms of δ without any assumption about ε, and write φ = φ0 + δφ1 + δ 2 φ2 + · · · ,
(9.3.7)
and then substitute in (9.3.3)–(9.3.5). The lowest-order term in (9.3.3) is φ0zz = 0,
(9.3.8)
which, with (9.3.6), yields φ0z ≡ 0, for all z, or φ0 = φ0 (x, y, t), which indicates that the horizontal velocity components are independent of the vertical coordinate z in lowest order. Consequently, we use the notation φ0x = u(x, y, t)
and φ0y = v(x, y, t).
(9.3.9ab)
The first- and second-order terms in (9.3.3) are given by φ0xx + φ0yy + φ1zz = 0, φ1xx + φ1yy + φ2zz = 0.
(9.3.10) (9.3.11)
Integrating (9.3.10) with respect to z and using (9.3.9ab) gives φ1z = −z(ux + vy ) + C(x, y, t),
(9.3.12)
where the arbitrary function C(x, y, t) becomes zero because of the bottom boundary condition (9.3.6). Integrating the resulting equation (9.3.12), again with respect to z and omitting the arbitrary constant, we obtain φ1 = −
z2 (ux + vy ), 2
(9.3.13)
so that φ1 = 0 at z = 0 and u and v are then the horizontal velocity components at the bottom boundary.
9.3 The Boussinesq and Korteweg–de Vries Equations
433
We next substitute (9.3.13) in (9.3.10), (9.3.11), and then integrate with condition (9.3.6) to determine the arbitrary function. Consequently, φ2z =
1 3 2 z ∇ u x + ∇2 v y , 6
φ2 =
1 4 2 z ∇ u x + ∇2 v y , 24 (9.3.14ab)
where ∇2 is the two-dimensional Laplacian. We next consider the free surface boundary conditions retaining all terms up to order δ, ε in (9.3.4), and δ 2 , ε2 , and δε in (9.3.5). It turns out that δ 1 φ0t − (utx + vty ) + η + ε u2 + v 2 = 0, (9.3.15) 2 2 δ 2 2
∇ u x + ∇2 v y . δ ηt + ε(uηx + vηy ) + (1 + εη)(ux + vy ) = 6 (9.3.16) Differentiating (9.3.15) first with respect to x and then with respect to y gives two equations: 1 ut + ε(uux + vvx ) + ηx − δ(utxx + vtxy ) = 0, 2 1 vt + ε(uuy + vvy ) + ηy − δ(utxy + vtyy ) = 0. 2
(9.3.17) (9.3.18)
Simplifying (9.3.16) yields δ 2 ηt + u(1 + εη) x + v(1 + εη) y = ∇ u x + ∇2 v y . 6
(9.3.19)
Equations (9.3.17)–(9.3.19) represent the nondimensional shallow water equations. Using the fact that φ0 is irrotational, that is, uy = vx and neglecting terms O(δ) in (9.3.17)–(9.3.19), we obtain the fundamental shallow water equations ut + ε(uux + vux ) + ηx = 0, vt + ε(uvx + vvy ) + ηy = 0, ηt + u(1 + εη) x + v(1 + εη) y = 0.
(9.3.20) (9.3.21) (9.3.22)
This system of three, coupled, nonlinear equations is closed and admits some interesting and useful solutions for u, v, and η. It is equivalent to the boundary-layer equations in fluid mechanics. Finally, it can be linearized when ε 1 to obtain the following dimensional equations: ut + gηx = 0,
vt + gηy = 0,
ηt + h(ux + vy ) = 0.
(9.3.23abc)
Eliminating u and v from these equations gives ηtt = c2 (ηxx + ηyy ).
(9.3.24)
434
9 Solitons and the Inverse Scattering Transform
This is a well-known two-dimensional wave equation. It corresponds to the √ nondispersive shallow water waves that propagate with constant velocity c = gh. This velocity is simply the linearized version of g(h + η). The wave equation has the simple d’Alembert solution representing plane progressive waves η(x, y, t) = f (k1 x + l1 y − κ1 ct) + g(k2 x + l2 y − κ2 ct),
(9.3.25)
where f and g are arbitrary functions and κ2r = (kr2 + lr2 ), r = 1, 2. We consider the one-dimensional case retaining both ε and δ order terms in (9.3.17)–(9.3.19) so that these equations reduce to the Boussinesq equations (1871a, 1871b, 1872) 1 ut + εuux + ηx − δutxx = 0, 2 1 ηt + u(1 + εη) x − δuxxx = 0. 6
(9.3.26) (9.3.27)
On the other hand, equations (9.3.20)–(9.3.22), expressed in dimensional form, are ut + uux + vuy + gHx = 0, vt + uvx + vvy + gHy = 0, Ht + (uH)x + (vH)y = 0,
(9.3.28) (9.3.29) (9.3.30)
where H = (h + η) is the total depth and Hx = ηx , since the depth h is constant. In particular, the one-dimensional version of the shallow water equations follows from (9.3.28)–(9.3.30) and is given by ut + uux + gHx = 0, Ht + (uH)x = 0.
(9.3.31) (9.3.32)
This system of approximate shallow water equations is analogous to the exact governing equations of gas dynamics for the case of a compressible flow involving only one space variable (Riabouchinsky 1932). √ It is convenient to rewrite these equations in terms of the wave speed c = gH by using dc = (g/2c) dH, so that they become ut + uux + 2ccx = 0, 2ct + cux + 2ucx = 0.
(9.3.33) (9.3.34)
The standard method of characteristics can easily be used to solve (9.3.33) and (9.3.34). Adding and subtracting these equations allows us to rewrite them in the characteristic form ∂ ∂ + (c + u) (u + 2c) = 0, (9.3.35) ∂t ∂x
9.3 The Boussinesq and Korteweg–de Vries Equations
∂ ∂ (u − 2c) = 0. + (c − u) ∂t ∂x
435
(9.3.36)
Equations (9.3.35), (9.3.36) show that u + 2c propagates in the positive x-direction with velocity c + u, and u − 2c travels in the negative x-direction with velocity c − u, that is, both u + 2c and u − 2c propagate in their respective directions with velocity c relative to the water. In other words,
u + 2c = const. on curves C+ on which dx dt = u + c, (9.3.37ab) = u − c, u − 2c = const. on curves C− on which dx dt where C+ and C− are characteristic curves of the system of partial differential equations (9.3.31), (9.3.32). A disturbance propagates along these characteristic curves at speed c relative to the flow speed. The quantities u ± 2c are called the Riemann invariants of the system, and a simple √ wave is propagating to the right into water of depth h, that is, u − 2c = c0 = gh. Then, the solution is given by 3 (9.3.38) u = f (ξ), x = ξ + c0 + u t, 2 where u(x, t) = f (x) at t = 0. However, we note that 3 ux = 1 − ux t f (ξ), 2 giving ux =
2f (ξ) . 2 + 3tf (ξ)
(9.3.39)
Thus, if f (ξ) is anywhere less than zero, ux tends to infinity as t → −2/(3f ). In terms of the free surface elevation, solution (9.3.38) implies that the wave profile progressively distorts itself, and, in fact, any forward-facing portion of such a wave continually steepens, or the higher parts of the wave tend to catch up with lower parts in front of them. Thus, all of these waves, carrying an increase of elevation, invariably break. The breaking of water waves on beaches is perhaps the most common and the most striking phenomenon in nature. An alternative system equivalent to the nonlinear evolution equations (9.3.26), (9.3.27) can be derived from the nonlinear shallow water theory, retaining both ε and δ order terms with δ < 1. This system is also known as the Boussinesq equations which, in dimensional variables, are given by (9.3.40) ηt + (h + η)u x = 0, 1 2 ut + uux + gηx = h uxxt . (9.3.41) 3 They describe the evolution of long water waves that move in both positive and negative x-directions. Eliminating η and neglecting terms smaller than O(ξ, δ) gives a single Boussinesq equation for u(x, t) in the form
436
9 Solitons and the Inverse Scattering Transform
utt − c2 uxx +
as
1 2 1 u xt = h2 uxxtt . 2 3
(9.3.42)
The linearized Boussinesq equation for u and η follows from (9.3.40) and (9.3.41) 2 2 1 2 ∂4 ∂ u 2 ∂ = 0. (9.3.43) −c − h 2 2 2 2 η ∂t ∂x 3 ∂x ∂t
This is in perfect agreement with the infinitesimal wave theory result expanded for small kh. Thus, the third derivative term in 9.3.41) may be identified with the frequency dispersion. Another equivalent version of the Boussinesq equation is given by 3 η2 1 + h2 ηxxxx . (9.3.44) ηtt − c2 ηxx = 2 h xx 3 There are several features of this equation. It is a nonlinear partial differential equation that incorporates the basic idea of nonlinearity and dispersion. Boussinesq obtained three invariant physical quantities, Q, E, and M , defined by 3 η Q= dx, η dx, E= η dx, M= −3 h −∞ −∞ −∞ (9.3.45) provided that η → 0 as |x| → ∞. Evidently, Q and E represent the volume and the energy of the solitary wave. The third quantity, M , is called the moment of instability, and the variational problem, δM = 0 with E fixed, leads to the unique solitarywave solution. Boussinesq also derived the results for the amplitude and volume of a solitary wave of given energy in the form 3 E 3/2 (9.3.46ab) a= , Q = 2hE 1/3 . 4 h
∞
∞
2
∞
ηx2
The former result shows that the amplitude of a solitary wave in a channel varies inversely as the channel depth h. The Boussinesq equation can then be written in the normalized form utt − uxx −
3 2 u xx − uxxxx = 0. 2
(9.3.47)
This particular form is of special interest because it admits inverse scattering formalism. Equation (9.3.47) has steady progressive wave solutions in the form u(x, t) = 4k2 f (X),
X = kx − ωt,
(9.3.48)
where the equation for f (X) can be integrated to obtain fXX = 6A + (4 − 6B)f − 6f 2 , where A is a constant of integration and
(9.3.49)
9.3 The Boussinesq and Korteweg–de Vries Equations
ω 2 = k 2 + k 4 (4 − 6B).
437
(9.3.50)
For the special case A = B = 0, a single solitary-wave solution is given by f (X) = sech2 (X − X0 ),
(9.3.51)
where X0 is a constant of integration. This result can be used to construct a solution for a series of solitary waves, spaced 2σ apart, in the form f (X) =
∞
sech2 (X − 2nσ).
(9.3.52)
n=−∞
This is a 2σ periodic function that satisfies (9.3.49) for certain values of A and B. We next assume that ε and δ are comparable, so that all terms O(ε, δ) in (9.3.17)– (9.3.19) can be retained. For the case of the two-dimensional wave motion (v ≡ 0 and ∂/∂y ≡ 0), these equations become 1 ut + ηx + εuux − δutxx = 0, 2 1 ηt + u(1 + εη) x − δuxxx = 0. 6
(9.3.53) (9.3.54)
We now seek steady progressive wave solutions traveling in the positive xdirection only, so that u = u(x − U t) and η = η(x − U t). With the terms of zero order in ε and δ and U = 1, we assume a solution of the form u = η + εP + δQ,
(9.3.55)
where P and Q are unknown functions to be determined. Consequently, equations (9.3.53), (9.3.54) become 1 (η + εP + δQ)t + ηx + εηηx − δηtxx = 0, 2 1 ηt + (1 + εη)(η + εP + δQ) x − δηxxx = 0. 6
(9.3.56) (9.3.57)
These equations must be consistent so that we stipulate for the zero order ηt = −ηx ,
1 P = − η2, 4
Q=
1 1 ηxx = − ηxt . 3 3
(9.3.58)
We use these results to rewrite both (9.3.56) and (9.3.57) with the assumption that ε and δ are of equal order and small enough for their products and squares to be ignored, so that the ratio (ε/δ) = (al2 /h3 ) is of order one. Consequently, we obtain a single equation for η(x, t) in the form 3 1 ηt + 1 + εη ηx + δηxxx = 0. (9.3.59) 2 6
438
9 Solitons and the Inverse Scattering Transform
This is now universally known as the Korteweg and de Vries equation as they discovered it in their 1895 seminal work. We point out that (ε/δ) = al2 /h3 is one of the fundamental parameters in the theory of nonlinear shallow water waves. Recently, Infeld (1980) considered three-dimensional generalizations of the Boussinesq and Korteweg–de Vries equations. Example 9.3.1 (Ion-Acoustic Waves and the KdV Equation). A high temperature plasma is a fully ionized gas consisting of electrons and ions that are governed by the equations of continuity and momentum combined with the classical Maxwell equations. Using the subscripts e and i for quantities related to electrons and ions and neglecting dissipation due to collisions, we write the following equations of motion for plasma. The equation of continuity is ∂nj + ∇ · (nj uj ) = 0, ∂t where nj (j = e, i) is the density and uj is the flow velocity. The equation of motion is ∂uj 1 + (uj · ∇)uj = −∇pj + nj qj E + (uj × B) . mj nj ∂t c
(9.3.60)
(9.3.61)
The Maxwell equations are given by ∇ · E = 4π(qi ni + qe ne ), ∇ · B = 0, (9.3.62ab) ∂B ∂E + c(∇ × E) = 0, − + c(∇ × B) = 4π(qi ni ui + qe ne ue ). ∂t ∂t (9.3.63ab) The equation of state is given by pj = nj Tj .
(9.3.64)
In the above equations, E is the electric field, B is the magnetic field, T is the product of the Boltzmann constant and the temperature, q and m are the charge and mass, respectively, and c is the speed of light. For an electrostatic wave, that is, for a one-dimensional longitudinal wave (∇ × E = 0, B = 0), we set qi = −qe ≡ e to obtain ∂nj ∂ + (nj uj ) = 0, ∂t ∂x 1 ∂ ∂uj e ∂uj E− + uj =± (Tj nj ), ∂t ∂x mj mj nj ∂x
(9.3.65) (9.3.66)
∂E ∂E = 4πe(ni − ne ), = 4πe(ne ue − ni ui ). (9.3.67ab) ∂x ∂t The motion of the cold ions in a hot electron gas with the classical Boltzmann distribution is governed by the simplified form of equations (9.3.60)–(9.3.64). In
9.3 The Boussinesq and Korteweg–de Vries Equations
439
this case, with me → 0 and Te = const. with Te Ti , it follows from equations (9.3.65)–(9.3.67ab) that eE =
Te ∂ne , ne ∂x
(9.3.68)
∂ ∂ni + (ni ui ) = 0, ∂t ∂x ∂ui e ∂ui 1 ∂ + ui = (ni Ti ), E− ∂t ∂x mi mi ni ∂x ∂E = 4πe(ni − ne ). ∂x
(9.3.69) (9.3.70) (9.3.71)
These are the governing equations for quantities ni , ui , E, and ne if Ti is a given constant or, by the adiabatic law, so that (ni Ti ) is proportional to nγi . Introducing the electrostatic potential φ, so that E = −φx , equation (9.3.69) gives the Boltz), where n0 is the number density of mann distribution for electrons ne = n0 exp( eφ Te electrons (or ions) in the unperturbed state. To examine the motion of the cold ions in a hot electron gas with the Boltzmann distribution, we use equations (9.3.68) and (9.3.69). It is convenient to nondimensionalize the physical quantities as follows: nj , n0 eφ , φ∗ = Te
n∗j =
uj , cs x x∗ = , λDe
u∗j =
E∗ = √
eE ωpe , me Te
t∗ = tωpi ,
1
where ωpe = (4πn0 /me ) 2 e is the electron plasma frequency, 1 1 −1 (Te /me ) 2 is the electron Debye length, ωpi = ωpe (me /mi ) 2 is the λDe = ωpe 1 ion plasma frequency, ni = n0 = const., and cs = λDe · ωpi = (Te /mi ) 2 . Consequently, omitting the asterisks, equations (9.3.69)–(9.3.71) reduce to nondimensional form ∂ni ∂ + (ni ui ) = 0, ∂t ∂x ∂ui ∂φ ∂ui + ui =− , ∂t ∂x ∂x ∂2φ = exp(φ) − ni . ∂x2
(9.3.72) (9.3.73) (9.3.74)
Substituting ni = 1 + ni exp[i(kx − ωt)] and (ui , φ) = (ui , φ ) exp[i(kx − ωt)] in the above equations gives the linear dispersion relation −1 ω 2 = k2 1 + k2 . (9.3.75) For long waves (k 2 1), equation (9.3.75) reduces to 1 2 ω ≈k 1− k . 2
(9.3.76)
440
9 Solitons and the Inverse Scattering Transform
Under the assumption of a quasi-neutral state, ni = ne = exp(φ) so that φ = log(ni ), the equation of motion (9.3.73) becomes ∂ui 1 ∂ni ∂ui + ui =− . ∂t ∂x ni ∂x
(9.3.77)
This is identical to the equation of motion for a gas with unit sound velocity (cs = 1). Physically, this means that electrons exert a pressure on ions through the electric field E = −φx , and hence, waves described by the above dispersion relation are called ion-acoustic waves. As the waves steepen by the inertia term, φxx (= ne −ni ) cannot be neglected. Consequently, a soliton is generated by the dispersion involved with the deviation from the quasi-neutral state (the second term of the dispersion relation). To obtain the soliton solution of the above system, we look for a stationary wave solution with ξ = x − ct and the boundary conditions ni → 1,
u2 → 0,
φ → 0,
∂φ → 0, ∂ξ
as |ξ| → ∞.
(9.3.78)
It turns out that φ(ξ, t) satisfies the nonlinear equation 2 1 1 ∂φ = exp(φ) + c c2 − 2φ 2 − c2 + 1 . 2 ∂ξ This gives the value of c2 when c2 =
∂φ ∂ξ
(9.3.79)
= 0 at φ = φm , so that
{1 − exp(φm )}2 . 2{exp(φm ) − (1 + φm )}
(9.3.80)
We assume that φm 1, so that c − 1 = δλ 1. Consequently, equation (9.3.79) reduces to the form 2 2 ∂φ = φ2 (3δc − φ). (9.3.81) ∂ξ 3 Integrating this equation gives a soliton solution of the form δc 2 φ(x, t) = 3(δc) sech (x − ct) . 2
(9.3.82)
This represents a KdV soliton when the amplitude is finite, but small. As the amplitude of the potential tends to φm = 12 c2 , that is, eφm = 12 mi c2 , the ions cannot pass the potential and so will be reflected. As a result, ion orbits will intersect, and hence, the cold ion approximation ceases to be valid. Accordingly, the maximum amplitude of the soliton is given by φm = 12 c2 , and, from equation (9.3.80), we obtain φm ∼ 1.3, and hence, c ∼ 1.6. This reveals that the maximum Mach number of the soliton is 1.6 as predicted by Sagdeev (1966). Example 9.3.2 (Derivation of the KdV Equation from the Euler Equations). We discussed this problem at the beginning of this section by using the Laplace equation
9.3 The Boussinesq and Korteweg–de Vries Equations
441
for the velocity potential under the assumption that δ = O(ε) as ε → 0. Here we follow Johnson (1997) to present another derivation of the KdV equation from the Euler equation in (1 + 1) dimensions. This approach can be generalized to derive higher dimensional KdV equations. We consider the problem of surface gravity waves which propagate in the positive x-direction over stationary water of constant depth. The associated Euler equations (2.7.67) and the continuity equation (2.7.68) in (1 + 1) dimensions are given by ut + ε(uux + wuz ) = −px , δ wt + ε(uwx + wwz ) = −pz , ux + wz = 0.
(9.3.83) (9.3.84) (9.3.85)
The free surface and bottom boundary conditions are obtained from (2.7.69), (2.7.70) in the form w = ηt + εuηx , p = η on z = 1 + εη, (9.3.86) w = 0 on z = 0. (9.3.87) √ It can easily be shown that, for any δ as ε → 0, there exists a region in the (x, t)space where there is a balance between nonlinearity and dispersion which leads to the KdV equation. The region of interest is defined by a scaling of the independent flow variables as δ δ x→ x and t → t, (9.3.88) ε ε √ for any ε and δ. In order to ensure consistency in the continuity equation, it is necessary to introduce a scaling of w by ε w. (9.3.89) w→ δ Consequently, the net effect of the scalings is to replace δ by ε in equations (9.3.83)–(9.3.87) so that they become ut + ε(uux + wuz ) = −px , ε wt + ε(uwx + wwz ) = −pz , ux + wz = 0, w = ηt + εuηx , w = 0 on z = 0.
and
p=η
(9.3.90) (9.3.91) (9.3.92) on z = 1 + εη,
(9.3.93) (9.3.94)
In the limit as ε → 0, the first-order approximation of equations (9.3.90) and (9.3.93) gives (9.3.95) η = p, 0 ≤ z ≤ 1, and ut + ηx = 0. It then follows from (9.3.92) that w = −zux which satisfies (9.3.94). The boundary condition (9.3.93) leads to ηt + ux = 0 on z = 1, which can be combined with (9.3.95) to obtain the linear wave equation
442
9 Solitons and the Inverse Scattering Transform
ηtt − ηxx = 0.
(9.3.96)
For waves propagating in the positive x-direction, we introduce the far-field variables ξ =x−t
and τ = εt,
(9.3.97)
so that ξ = O(1) and τ = O(1) give the far-field region of the problem. This is the region where nonlinearity balances the dispersion to produce the KdV equation. With the choice of the transformations (9.3.97), equations (9.3.90)–(9.3.94) can be rewritten in the form −uξ + ε(uτ + uuξ + wuz ) = −pξ , ε −wξ + ε(wτ + uwξ + wwz ) = −pz , uξ + wz = 0, w = −ηξ + ε(ητ + uηξ ), w = 0 on z = 0.
p=η
on z = 1 + εη,
(9.3.98) (9.3.99) (9.3.100) (9.3.101) (9.3.102)
We seek an asymptotic series expansion of the solutions of the system (9.3.98)– (9.3.102) in the form ∞
η(ξ, τ, ε) =
n
ε ηn (ξ, τ ),
q(ξ, τ, z; ε) =
n=0
∞
εn qn (ξ, τ, z),
(9.3.103)
n=0
where q (and the corresponding qn ) denotes each of the variables u, w, and p. Consequently, the leading-order problem is given by u0ξ = p0ξ , p0z = 0, u0ξ + w0z = 0, p0 = η0 , w + η0ξ = 0 on z = 1, w = 0 on z = 0.
(9.3.104) (9.3.105) (9.3.106)
These leading-order equations give p 0 = η0 ,
u0 = η0 ,
w0 + zη0ξ = 0,
0 ≤ z ≤ 1,
(9.3.107)
with u0 = 0 whenever η0 = 0. The boundary condition on w0 at z = 1 is automatically satisfied. Using the Taylor series expansion of u, w, and p about z = 1, the two surface boundary conditions on z = 1 + εη are rewritten on z = 1, and hence, take the form on z = 1, (9.3.108) p0 + εη0 p0z + εp1 = η0 + εη1 + O ε2 2 w0 + εη0 w0z + εw1 = −η0ξ − εη1ξ + ε(η0τ + u0 η0ξ ) + O ε on z = 1. (9.3.109) These conditions are to be used together with (9.3.98), (9.3.99), and (9.3.102).
9.3 The Boussinesq and Korteweg–de Vries Equations
443
The equations in the next order are given by −u1ξ + u0τ + u0 u0ξ + w0 u0z = −p1ξ ,
(9.3.110)
p1z = w0ξ ,
(9.3.111)
u1ξ + w1z = 0, p1 + η0 p0 = η1 ,
w1 + η0 + w0z = −η1ξ + η0τ + u0 η0ξ
on z = 1, (9.3.112)
w1 = 0 on z = 0.
(9.3.113)
Noting that u0z = 0,
p0z = 0,
we obtain p1 =
and
w0z = −η0ξ ,
1 1 − z 2 η0ξξ + η1 , 2
(9.3.114) (9.3.115)
and therefore, w1z = −u1ξ = −p1ξ − u0τ − u0 u0ξ 1 = − η1ξ + 1 − z 2 η0ξξξ + η0τ + η0 η0ξ . 2
(9.3.116)
Finally, we find 1 1 w1 = − η1ξ + η0τ + η0 + η0ξ + η0ξξξ z + z 3 η0ξξξ , 2 6 which satisfies the bottom boundary condition on z = 0. The free surface boundary condition on z = 1 gives 1 1 (w1 )z=1 = − η1ξ + η0τ + η0 η0ξ + η0ξξξ + η0ξξξ 2 6 = −η1ξ + η0τ + 2η0 η0ξ ,
(9.3.117)
(9.3.118)
which yields the equation for η0 (ξ, τ ) in the form 3 1 η0τ + η0 η0ξ + η0ξξξ = 0. 2 6
(9.3.119)
This is the Korteweg–de Vries (KdV) equation, which describes nonlinear plane gravity waves propagating in the x-direction. It will be described in Section 9.7 that the exact solution of the general initial-value problem for the KdV equation can be obtained, provided the initial data decay sufficiently rapidly as |ξ| → ∞. We may raise the question of whether the asymptotic expansion for η (and hence, for the other flow variables) is uniformly valid as |ξ| → ∞ and as τ → ∞. For the case of τ → ∞, this question is difficult to answer because the equations for ηn (n ≥ 1) are not easy to solve. However, all the available numerical evidence suggests that the asymptotic expansion of η is indeed uniformly valid as τ → ∞ (at least for solutions which
444
9 Solitons and the Inverse Scattering Transform
satisfy η → 0 as |ξ| → ∞). From a physical point of view, if the waves are allowed to propagate indefinitely, then other physical effects cannot be neglected. In the case of real water waves, the most common physical effects include viscous damping. In practice, the viscous damping seems to be sufficiently weak to allow the dispersive and nonlinear effects to dominate before the waves eventually decay completely. It is well known that the KdV equation describes nonlinear plane waves which propagate only in the x-direction. However, there are many physical situations in which waves move on a two-dimensional surface. So it is natural to include both xand y-directions with the appropriate balance of dispersion and nonlinearity. One of the simplest examples is the classical two-dimensional linear wave equation utt = c2 (uxx + uyy ), which describes the propagation of long waves. This equation has a solution in the form u(x, t) = a exp i(ωt − k · x) ,
(9.3.120)
(9.3.121)
where a is the wave amplitude, ω is the frequency, x = (x, y), and the wavenumber vector is k = (k, ). The dispersion relation is given by ω 2 = c2 k2 + 2 . (9.3.122) For waves that propagate predominantly in the x-direction, the wavenumber component becomes small so that the approximate phase velocity is given by 1 2 as → 0. (9.3.123) cp = c 1 + 2 k2 Example 9.3.3 (The Kadomtsev–Petviashvili (KP) Equation or Two-Dimensional KdV Equation). It follows from (9.3.123) that the phase velocity suffers from a small correction provided by the wavenumber component in the y-direction. In order to ensure that this small correction is the√same order as the dispersion and nonlinearity, it it necessary to require = O( ε) or 2 = O(ε). This requirement can be incorporated in the governing equations by a scaling of the flow variables as √ √ (9.3.124) y → εy and v → εv, and using the same far-field transformations as (9.3.97). Consequently, equations (2.7.67)–(2.7.71) with the parameters δ replaced by ε reduce to the form −uξ + ε(uτ + uuξ + εvuy + wuz ) = −pξ , −vξ + ε(vτ + uvξ + εvvy + wvz ) = −py , −ε −wξ + ε(wτ + uwξ + εvwy + wwz ) = −pz , −uξ + εvy + wz = 0,
(9.3.125) (9.3.126) (9.3.127) (9.3.128)
9.3 The Boussinesq and Korteweg–de Vries Equations
w = −ηξ + ε(ητ + uuξ + εvηy ), w = 0 on z = 0.
and
p=η
on z = 1 + εη,
445
(9.3.129) (9.3.130)
We seek the same asymptotic series solution (9.3.103) valid as ε → 0 without any change of the leading order problem except that the flow variables involve y so that p0 = η0 , v0ξ = η0y .
u0 = η0 ,
w0 = −zη0ξ ,
0 ≤ z ≤ 1,
(9.3.131) (9.3.132)
At the next order, the only difference from Example 9.3.2 is in the continuity equation which becomes (9.3.133) w1z = −u1ξ − v0y . This change leads to the following equation: 1 1 w1 = − η1ξ + η0τ + η0 η0ξ + η0ξξξ + v0y z + z 3 η0ξξξ . 2 6
(9.3.134)
Using equation (9.3.132), the final result is the equation for the leading-order representation of the surface wave in the form 1 2η0τ + 3η0 η0ξ + η0ξξ + v0y = 0. 3
(9.3.135)
Consequently, differentiating (9.3.135) with respect to ξ and replacing v0ξ by η0y give the evolution equation for η0 (ξ, τ, y) in the form 1 2η0τ + 3η0 η0ξ + η0ξξξ + η0yy = 0. (9.3.136) 3 ξ This is the two-dimensional or, more precisely, the (1 + 2)-dimensional KdV equation. Obviously, when there is no y-dependence, (9.3.136) reduces to the KdV equation (9.3.119). The two-dimensional KdV equation is also known as the Kadomtsev–Petviashvili (KP) equation (Kadomtsev and Petviashvili 1970). This is another very special completely integrable equation, and it has an exact analytical solution that represents obliquely crossing nonlinear waves. Physically, any number of waves cross obliquely and interact nonlinearly. In particular, the nonlinear interaction of three waves leads to a resonance phenomenon. Such an interaction becomes more pronounced, leading to a strongly nonlinear interaction as the wave configuration is more nearly that of parallel waves. This situation can be interpreted as one in which the waves interact nonlinearly over a large distance so that a strong distortion is produced among these waves. The reader is referred to Johnson’s (1997) book for more detailed information on oblique interaction of waves. We now consider the Euler equations (2.7.4)–(2.7.6) and the continuity equation (2.7.2) in cylindrical polar coordinates (r, θ, z). It is convenient to use the nondimensional flow variables, parameters, and scaled variables similar to those defined
446
9 Solitons and the Inverse Scattering Transform
by (2.7.56), (2.7.57), and (2.7.66) with r = r∗ where r∗ is a nondimensional variable so that the polar form of Euler equations (2.7.4)–(2.7.6), continuity equation (2.7.2), and the boundary conditions (2.7.64), (2.7.65) in nondimensional form become ∂p Dv εuv 1 ∂p Dw ∂p Du εv 2 − =− , + =− , δ =− , Dt r ∂r Dt r r ∂θ Dt ∂z 1 ∂ 1 ∂v ∂w (ru) + + = 0, r ∂r ∂z r ∂θ v w = ηt + ε uηr + ηθ , and p = η on z = 1 + εη, r w = 0 on z = 0, where
D ∂ ∂ v ∂ ∂ ≡ +ε u + +w , Dt ∂t ∂r r ∂θ ∂z
(9.3.137) (9.3.138) (9.3.139) (9.3.140)
(9.3.141)
and ε and δ are defined by (2.7.58). ∂ ≡ 0), the governing equations For the case of axisymmetric wave motions ( ∂θ become ut + ε(uur + wuz ) = −pr , δ wt + ε(uwr + wwz ) = −pz , u ur + + wz = 0, r w = ηt + εuηr , and p = η on z = 1 + εη, w = 0 on z = 0.
(9.3.142) (9.3.143) (9.3.144) (9.3.145) (9.3.146)
In the limit as ε → 0, the linearized version of equations (9.3.137)–(9.3.141) become 1 vt = − pθ , δwt = −pz , r 1 ∂v ∂w 1 ∂ (ru) + + = 0, r ∂r r ∂θ ∂z w = ηt , and p = η on z = 1,
ut = −pr ,
w=0
on z = 0,
(9.3.147) (9.3.148) (9.3.149) (9.3.150)
For long waves (δ → 0), equations (9.3.147)–(9.3.150) lead to the classical wave equation 1 1 (9.3.151) ηtt = ηrr + ηr + 2 ηθθ . r r For axisymmetric surface waves, the wave equation (9.3.151) becomes 1 ηtt = ηrr + ηr . r This can be solved by using the Hankel transform (see Example 1.10.3).
(9.3.152)
9.3 The Boussinesq and Korteweg–de Vries Equations
447
It is convenient to introduce the characteristic variable ξ = r − t for outgoing waves and R = αr so that α → 0 corresponds to large radius r. In other words, R = O(1), as α → 0, gives r → ∞. The equation (9.3.152) reduces to the following form 1 1 (9.3.153) 2ηξR + ηξ = α ηRR + ηR = 0. R R In the limit as α → 0, it follows that √ Rηξ = g(ξ),
(9.3.154)
where g(ξ) is an arbitrary function of ξ. For outgoing waves, the correct solution takes the form for α → 0 (r → ∞), 1 1 η= √ g(ξ) dξ = √ f (ξ), (9.3.155) R R where η = 0 when f = 0. This shows that the amplitude of waves decays as the radius r → ∞(R → ∞). This behavior is totally different from the derivation of the KdV equation where the amplitude is uniformly O(ε). In the present axisymmetric case, the amplitude decreases as the radius r increases so that there is no far-field region where the balance between nonlinearity and dispersion occurs. In other words, the amplitude is so small that nonlinear terms play no role at the leading order. However, there exists a scaling of the flow variables which leads to the axisymmetric (concentric) KdV equation as shown by Johnson (1997). We recall the axisymmetric Euler equations of motion (9.3.142)–(9.3.146) and introduce scalings in terms of large radial variable R, ξ=
ε2 (r − t) δ
and R =
ε6 r. δ2
We next apply the transformations of the flow variables ε3 ∗ ∗ ∗ ε2 ∗ (η, p, u, w) = η ,p ,u , w , δ δ
(9.3.156)
(9.3.157)
where large distance/time is measured by the scale (δ 2 /ε6 ) so that 3 2 − 12 ε δ = , ε6 δ which represents the scale of the amplitude of the waves. It is important to point out that the original wave amplitude parameter ε can now be interpreted based on the amplitude of the wave for r = O(1) and t = O(1). Consequently, the governing equations (9.3.142)–(9.3.146) become, dropping the asterisks in the variables, −u ξ + αpR ), ξ + α(uuξ + wuz + αuuR ) = −(p α −wξ + α(uwξ + wwz + αuwR ) = −pz ,
(9.3.158) (9.3.159)
448
9 Solitons and the Inverse Scattering Transform
1 uξ + wz + α uR + u = 0, R w = −ηξ + α(uηξ + αuηR ), p = η w = 0 on z = 0,
(9.3.160) on z = 1 + αη,
(9.3.161) (9.3.162)
where α = (δ −1 ε4 ) is a new parameter. These equations are similar in structure to those discussed above with parameter ε, which is now replaced by α in (9.3.158)– (9.3.162) so that the limit as α → 0 is required. This requirement is satisfied (for example, ε → 0 with δ fixed), and the scaling introduced by (9.3.156) describes the region where the appropriate balance occurs so that the wave amplitude in this region is O(α). We now seek an asymptotic series solution in the form q(ξ, R, z) =
∞
αn qn (ξ, R, z),
(9.3.163)
n=0
where q represents each of η, u, w, and p. In the leading order, we obtain the familiar equations p0 = η 0 ,
u0 = η0 ,
w0 = −zη0ξ ,
0 ≤ z ≤ 1.
It follows from the continuity equation (9.3.160) that 1 w1z = −u1z − u0R + u0 . R
(9.3.164)
(9.3.165)
Without any more algebraic calculation, it turns out that η0 (ξ, R) satisfies the nonlinear evolution equation 2η0R +
1 1 η0 + 3η0 η0ξ + η0ξξξ = 0. R 3
(9.3.166)
This is usually referred to as the axisymmetric (concentric) KdV equation which includes a new term R−1 η0 . We may use the large time variable τ = (δ−2 ε6 )t so that R = τ + αξ ≈ τ (see Johnson 1997 or 1980). Example 9.3.4 (Derivation of Johnson’s Evolution Equation). Johnson (1980) derived a new concentric KdV equation which incorporates weak dependence on the √ angular coordinate θ. In the derivation of KP equation (9.3.136), ε was chosen as the scaling parameter in the y-direction. √ Similarly, the appropriate scaling on the angular variable θ may be chosen as α. In the derivation of the concentric KdV equation (9.3.166), the parameter α plays the role of ε which is used as the small parameter in the asymptotic solution of the KdV equation. Following the work of Johnson (1980), we choose the variables ξ and R defined by (9.3.156) and the scaled θ variable as √ θ = α θ∗ = δ −1 ε2 , (9.3.167)
9.3 The Boussinesq and Korteweg–de Vries Equations
449
which introduces a small angular deviation from purely concentric effects. We also use the scaled velocity component in the θ-direction as 1 (9.3.168) v = δδ − 2 ε5 v ∗ , so that the scalings on u = φr and v = 1r φθ are found to be consistent. Invoking (9.3.156), (9.3.157) combined with (9.3.167), (9.3.168), and dropping the asterisk from all variables the governing equations (9.3.137)–(9.3.140) can be expressed in the following form: α3 v − v = −(pξ + αpR ), (9.3.169) −uξ + α uuξ + wuz + α uuR + uθ R R 1 1 α2 uv = − pθ , −vξ + α uvξ + wvz + α uvR + vvθ (9.3.170) + R R R α = −pz , (9.3.171) α −wξ + α uwξ + wwz + αuwR + vwθ R 1 uξ + wz + α uR + (u + vθ ) = 0, (9.3.172) R 1 w = −ηξ + α uηξ + α uηR + vηθ , and p = η, on z = 1 + αη, R (9.3.173) w = 0 on z = 0. (9.3.174) An asymptotic solution similar to that of (9.3.163) gives the same result (9.3.164) at the leading order with (9.3.175) v0ξ = R−1 η0θ . At the next order, the continuity equation (9.3.172) produces a different result in the form 1 w1z = − u1ξ + u0R + (u0 + v0θ ) . (9.3.176) R Omitting some algebraic calculations, it turns out that (9.3.176) leads to a KdV type equation 2η0R +
1 1 1 η0 + 3η0 η0ξ + η0ξξξ + v0θ = 0. R 3 R
(9.3.177)
Eliminating v0θ from (9.3.177) with the help of (9.3.175) gives the equation in the form 1 1 1 2η0R + η0 + 3η0 η0ξ + η0ξξξ + 2 η0θθ = 0. (9.3.178) R 3 R ξ This is known as Johnson’s equation, as it was first derived by Johnson (1980). In the absence of the θ-dependence, equation (9.3.178) reduces to the concentric KdV equation (9.3.166).
450
9 Solitons and the Inverse Scattering Transform
Example 9.3.5 (Transformation of the KP Equation to the Concentric KdV Equation). Replacing τ by t and ξ by x, we rewrite the KP equation (9.3.136) in the form 1 + ηyy = 0. (9.3.179) 2ηt + 3ηηx + ηxxx 3 x We apply the transformations η = h(x∗ , t),
x∗ = x +
1 y2 2 t
(9.3.180)
to equation (9.3.179) to obtain, dropping the asterisk, 2 2 1 1 y y 2ht − hx + 3hhx + hxxx + hx + hxx = 0. t 3 t t x
(9.3.181)
Integrating this equation with respect to x and using the decay condition for x → ∞ gives the concentric KdV equation for h(x, t) 1 1 2ht + h + 3hhx + hxxx = 0. t 3
(9.3.182)
This equation is similar to (9.3.166) in which R, ξ are replaced by t and x, respectively. Example 9.3.6 (Transformation of Johnson’s Equation to the KdV Equation). We recall Johnson’s equation (9.3.178) 1 1 1 (9.3.183) 2ηR + η + 3ηηξ + ηξξξ + 2 ηθθ = 0. R 3 R ξ We use cylindrical polar coordinates (x = r cos θ, θ → 0. We write 1 y2 r−t∼x 1+ − t = x 1+ 2 x2 so that
y = r sin θ) in the limit as 1 2 θ 2
−t
δ 1 2 1 ∗2 x − t ∼ r − t − rθ = 2 ξ − Rθ . 2 ε 2
We next apply the transformations h = η(ζ, R),
1 ζ = ξ − Rθ ∗2 2
to equation (9.3.178) and integrate the resulting equation with respect to ξ to obtain the equation 1 (9.3.184) 2hR + 3hhζ + ηζζζ = 0. 3 This is the usual KdV equation where R = τ + αξ may be replaced by τ as α → 0 and R may be interpreted as the time variable.
9.3 The Boussinesq and Korteweg–de Vries Equations
451
Example 9.3.7 (Derivation of the Boussinesq Equation from the Euler Equations). We consider another example of weakly nonlinear and weakly dispersive waves which propagate in both the positive and negative x-directions. We use the Euler equations (9.3.83)–(9.3.87) with the scaling which replaces δ by ε so that ut + ε(uux + wuz ) = −px , ε wt + ε(uwx + wwz ) = −pz , ux + wz = 0, w = ηt + εηηx ,
and
p=η
(9.3.185) (9.3.186) (9.3.187) on z = 1 + εη,
on z = 0.
w=0
(9.3.188) (9.3.189)
We seek as usual an asymptotic expansion of solutions in integral powers of ε as ε → 0. At the leading order O(1), we obtain p0 = η0 ,
u0t = −η0x ,
w0 = −zu0x ,
u0x = −η0t ,
0 ≤ z ≤ 1. (9.3.190)
Consequently, η0 satisfies the classical wave equation η0tt − η0xx = 0.
(9.3.191)
In this problem, we seek solutions that are valid for x = O(1) and t = O(1). At the next order O(ε), we find p1 = −
1 1 − z 2 u0xt + η1 2
so that u1t + u0 u0x =
1 1 − z 2 u0xxt − η1x . 2
(9.3.192)
(9.3.193)
And the equation w1z = −u1x
(9.3.194)
1 1 − z 2 u0xxxt + (u0 u0x )x . 2
(9.3.195)
leads to the following form: w1zt = −u1xt = η1xx −
Integrating this equation with respect to z gives 1 1 w1t = z (u0 u0x )x + η1xx − u0xxxt + z 3 u0xxxt . 2 6
(9.3.196)
This satisfies w1t = 0 or w1 = 0 on z = 0. Differentiating with respect to t, the free surface boundary condition on z = 1 yields (w1 + η0 w0z )t = (η1t + u0 η0x )t .
(9.3.197)
452
9 Solitons and the Inverse Scattering Transform
Thus, 1 (u0 u0x )x + η1xx − u0xxxt − (η0 u0x )t = η1tt + (u0 η0x )t , 3 which becomes, by using (9.3.190), 1 1 − η0xxx = 0. η1tt − η1xx − u20 + η02 2 3 xx
(9.3.198)
(9.3.199)
Combining (9.3.191) and (9.3.199) gives a single equation for η = η0 + εη1 + O(ε2 ) in the form 2 x 1 2 ε ηt dx − ηxxxx = O ε2 , ηtt − ηxx − ε η + (9.3.200) 2 3 −∞ xx where we have used
u0 = −
x −∞
η0 dx
under the assumption that u0 → 0 as x → ∞. Equation (9.3.200) or equation (9.3.200) with zero on the right-hand side is referred to as the Boussinesq (1871a, 1871b) equation. This equation describes weakly nondispersive waves which propagate in both the positive and negative x-directions. Finally, invoking the transformation η ∗ = η − εη 2 and the definition x ∗ x =x+ε η(x, t, ε) dx, −∞
the equation for η ∗ (x∗ , t, ε) takes the form, dropping the asterisk, 3 1 ηtt − ηxx − ε η 2 xx − εηxxxx = O ε2 . 2 3
(9.3.201)
This equation or this equation with zero on the right-hand side is the usual form of the Boussinesq equation, which is found to be completely integrable for any positive ε. Using the transformations ε ε 2 η → − η, x→ x, and t → t, ε 3 3 equation (9.3.201) reduces to the standard version of the Boussinesq equation as (9.3.202) ηtt − ηxx + 3 η 2 xx − ηxxxx = 0. Example 9.3.8 (Derivation of the KdV Equations for Gravity–Capillary Waves). The two-dimensional exact nonlinear gravity-capillary waves in water of finite depth h are governed by the following equations for the velocity potential φ(x, z, t) and the free surface elevation η(x, t): ∇2 φ = φxx + φzz = 0, 0 < z < h + η, ηt + φx ηx − φz = 0 on z = h + η,
(9.3.203) (9.3.204)
9.3 The Boussinesq and Korteweg–de Vries Equations
T ηxx 1 2 φx + φ2z − =B 2 ρ (1 + ηx2 )3/2 φz = 0 on z = 0,
φt + gη +
on z = h + η,
453
(9.3.205) (9.3.206)
where the origin of coordinates is at z = 0, T is the surface tension, ρ is the constant density of water, and B is a constant. In terms of nondimensional flow variables defined by (9.3.1) and two fundamental parameters ε and δ to characterize the nonlinear shallow gravity–capillary waves, equations (9.3.203)–(9.3.206) can be written, dropping the asterisks, in the nondimensional form: δφxx + φzz δ(ηt + εφx ηx ) − φz ε 2 1 2 τ δηxx φt + η + φ + φ − 2 x δ z (1 + ε2 δηx2 )3/2 φz
= 0, 0 < z < 1 + εη, = 0 on z = 1 + εη,
(9.3.207) (9.3.208)
= 0 on z = 1 + εη,
(9.3.209)
= 0 on z = 0,
(9.3.210)
T where τ = ( gρh 2 ) is the Bond number. Following Vanden-Broeck (2010), we seek series expansion of φ(x, z, t) in powers of z in the form ∞ z n fn (x, t). (9.3.211) φ(x, z, t) = n=0
Substituting (9.3.211) into (9.3.207) and (9.3.210) gives for n = 1, 2, 3, . . . , δ ∂ 2 fn−2 fn = − , n = 2, 4, 6, . . . . n(n − 1) ∂x2 fn = 0,
(9.3.212) (9.3.213)
Solving (9.3.213) recursively yields φ=
∞
(−1)n δ n
n=0
z n ∂ 2n f , (2n)! ∂x2n
where f = f0 .
(9.3.214)
Substituting (9.3.214) into the free surface conditions (9.3.208)–(9.3.209) yields ε 1 ηt + (1 + εη)fx x − δ (1 + εη)3 fxxxx + (1 + εη)2 ηx fxxx 6 2 2 + O δ = 0, (9.3.215) τ δη 1 2 δ xx 2 − ft + η + εfx − (1 + εη)2 fxxt + εfx fxxx − εfxx 2 2 (1 + ε2 δηx2 )3/2 + O δ 2 = 0, (9.3.216) Dropping all terms of order δ and then differentiating (9.3.216) with respect to x gives
454
9 Solitons and the Inverse Scattering Transform
ηt + (1 + εη)fx x = 0, fxt + ηx + εfx fxx − (τ δ)ηxxx = 0.
(9.3.217) (9.3.218)
Equations (9.3.217)–(9.3.218) represent the shallow water equations for the gravity– capillary waves. These define a hyperbolic system of partial differential equations which does not admit traveling wave solutions. The major result of this analysis is that the KdV equation for gravity–capillary waves can be obtained from (9.3.217)–(9.3.218) for waves moving to the right. To the lowest order in ε and δ, that is, to the order ε0 and δ 0 , these two equations become ηt + fxx = 0
and
fxt + ηx = 0.
(9.3.219)
Thus, the solution moving to the right is η = fxx ,
ηt + ηx = 0.
We assume solutions to order ε and δ in the form fx = η + εP + δQ + O ε2 + δ 2 ,
(9.3.220)
(9.3.221)
where P and Q are unknown functions of η and its derivatives. Consequently, equations (9.3.115)–(9.3.116) reduce to 1 ηt + ηx + ε(Px + 2ηηx ) + δ Qx − ηxxx = 0, (9.3.222) 6 1 ηt + ηx + ε(Pt + ηηx ) + δ Qt − ηxxt − δτ ηxxx = 0. (9.3.223) 2 Since ηt = −ηx + O(ε, δ), replacing the t-derivatives by minus the x-derivatives in the first order terms in (9.3.223) gives 1 (9.3.224) ηt + ηx + ε(−Px + ηηx ) + δ −Qx + ηxxx − δτ ηxxx = 0. 2 Equations (9.3.222)–(9.3.224) must be consistent so that the coefficients of ε and δ in these equations are the same. Consequently, 2 − τ ηxxx . 2Px = −ηηx , 2Qx = (9.3.225) 3 Integrating (9.3.225) with respect to x gives 1 P = − η2 4
and
Q=
1 1 − τ ηxx . 3 2
(9.3.226)
Substituting the values for Px and Qx into (9.3.224) gives the KdV equation for gravity–capillary waves in the form
9.3 The Boussinesq and Korteweg–de Vries Equations
3 1 ηt + 1 + εη ηx + (1 − 3τ )δηxxx = 0. 2 6
455
(9.3.227)
In terms of dimensional variables, the KdV equation (9.3.227) reduces to the form 3 ch2 (9.3.228) η ηx + (1 − 3τ )ηxxx = 0, ηt + c 1 + 2h 6 √ where c = gh is the shallow water wave velocity and the total depth of water is h + η. It is important to point out that, in the absence of surface tension (T = 0, τ = 0), equation (9.3.228) reduces to the celebrated KdV equation (9.3.1). We next seek a traveling wave solution of (9.3.228) in the form η(X) = hζ(X),
X = x − U t,
(9.3.229)
where η(X) represents a wave moving to the right with constant velocity U . Substituting (9.3.229) into (9.3.228) gives the ordinary differential equation 3c 1 hζζ + ch2 (1 − 3τ )hζ = 0, h(c − U )ζ + (9.3.230) 2 6 where the primes denote the derivatives with respect to X. Integrating (9.3.230) with respect to X yields h2 3 U ζ + ζ 2 + (1 − 3τ )ζ = A, 1− c 4 6 where A is a constant of integration. Multiplying (9.3.231) by ζ and integrating again gives U 2 h2 3 ζ +2 1− ζ + (1 − 3τ )ζ 2 = 4Aζ + B, c 3
(9.3.231)
(9.3.232)
where B is another constant of integration. We consider a special case when ζ and its derivatives tend to zero at infinity (X → ±∞). Hence, A = B = 0 so that (9.3.232) gives
dζ dX
2 =
3 2 ζ (a − ζ), h21
where a=2
h21 = h2 (1 − 3τ ),
U −1 . c
(9.3.233)
(9.3.234)
The right-hand side of (9.3.233) vanishes at ζ = 0 and ζ = a, and the exact solution of (9.3.233) can be written in the integral form X − X0 =
h21 3
12 0
ζ
dζ √ , ζ a−ζ
(9.3.235)
456
9 Solitons and the Inverse Scattering Transform
which is, by substitution ζ = a sech2 θ, X − X0 =
h21 3a
12 (9.3.236)
θ,
where X0 is an integrating constant, Thus, the solution is ζ(X) = a sech
2
3a 4h21
12
(X − X0 ) .
(9.3.237)
The solution ζ(X) increases from ζ = 0 as X → −∞ so that it attains a maximum value ζ = ζmax = a at X = 0, and then decreases symmetrically to ζ = 0 as X → ∞. These features imply that X0 = 0 so that 2
ζ(X) = a sech
3a 4h21
12
(9.3.238)
X .
When τ < 13 and a > 0, (9.3.238) represents a wave of elevation traveling with the speed U given by (9.3.234), that is, a U =c 1+ > c. (9.3.239) 2 When τ > 13 and a < 0, the solution ζ(X) in(9.3.238) is a wave of depression moving with speed U < c. Therefore, the final solution is 2
η(x, t) = η0 sech
3η0 4h3 (1 − 3τ )
12
(x − U t) ,
(9.3.240)
where the range in η is η0 = ah1 . As explained in Section 9.4, η(x, t) represents capillary solitary wave solution of the KdV equation (9.3.227) for any positive constant η0 . Since η(X) > 0 for all X, the solution is a wave of elevation which is symmetrical about X = 0. It propagates in the medium without change of shape with ve3η0 − 12 locity U which is directly proportional to the amplitude η0 . The width, [ 4h3 (1−3τ )] √ of the solution is inversely proportional to η0 . It is important to point out that, when τ = 13 , solution (9.3.238) shows that dζ ( dX ) becomes unbounded. This is due to the fact that the linear dispersive term in (9.3.238) disappears when τ = 13 . Furthermore, equation (9.3.228) is derived under the assumption that ε and δ are both small and of the same order of magnitude, that is, ε = δ = α, where 0 < α 1. This assumption has successfully been used in the expansion procedure to derive (9.3.228). However, to derive the evolution equation valid near τ = 13 , it is necessary to include higher order dispersive terms. This can be achieved by making use of ε = α2 and δ = α and using the expansion near 13 in the form 1 τ = + ατ1 + α2 τ2 + · · · . (9.3.241) 3
9.3 The Boussinesq and Korteweg–de Vries Equations
457
Fig. 9.4 A solitary wave with monotonic decay to a constant level.
Fig. 9.5 (a) An elevation solitary wave with a decaying oscillatory tail. (b) A depression solitary wave with a decaying oscillatory tail.
Invoking such expansion procedure, the fifth order KdV equation for gravity– capillary waves can be derived in the form 3 ch2 ch4 η ηx + (1 − 3τ )ηxxx + ηxxxxx = 0. (9.3.242) ηt + c 1 + 2h 6 90 We close this section by adding the following comments based on VandenBroeck’s (2010) computational results of gravity–capillary solitary waves. When 0 < τ < 13 , there are no fully nonlinear solutions of small amplitude consistent with (9.3.240). However, there are fully nonlinear solutions characterized by infinite train of ripples in the far field. Such waves are often referred to as generalized solitary waves with a series of ripples of constant amplitude in the far field. The above analysis further reveals that there are three different kinds of capillary solitary waves. The first kind is a solitary wave with a free-surface elevation that tends monotonically to a constant level in the far field as shown in Figure 9.4. The solution (9.3.238) or (9.3.240) is an example of this kind. The solitary wave of the second kind represents a free-surface elevation (or depression) wave profile which tends to a constant level in the far field with a decaying oscillatory tail as shown in Figures 9.5(a) and 9.5(b).
458
9 Solitons and the Inverse Scattering Transform
Fig. 9.6 A solitary wave with oscillations of constant amplitude.
The third kind of solitary wave represents a free-surface wave profile that does not tend to a constant level, but is characterized by oscillations of constant amplitude in the far field as shown in Figure 9.6. This is often called the generalized solitary wave. Such waves occur when gravity and surface tension are included. The existence of solitary waves as shown in Figures 9.5(a) and 9.5(b) is related to the minimum in Figure 2.4 in Chapter 2 for τ < 13 . The profiles in Figures 9.5(a) and 9.5(b) look like waves of slowly varying amplitude. However, those waves cannot be expected to be steady unless the phase and group velocities are equal. This is exactly what happens at the minimum in Figure 2.4. This reveals an intuitive explanation why the branches of solitary waves decaying tails bifurcate from the minimum value of the Figure 2.4. Numerical computations of several authors including Vanden-Broeck (2010), Vanden-Broeck and Dias (1992), and others confirmed the existence of such waves. On the other hand, numerical results of Vanden-Broeck (2010) and Champneys et al. (2002) show that the multiple branches of periodic solutions approach generalized solitary waves (see Figure 9.6) as λh = 2π kh → ∞. In other words, moving from one branch to the next includes two crests or two troughs. In the limit, as λ h =→ ∞, this produces a generalized solitary wave with an infinite number of crests and troughs. Thus, these ripples in the tail of the generalized solitary waves seem to be physically unrealistic because they occur on both sides, and hence, they do not satisfy the radiation conditions at infinity. This unrealistic feature has been rigorously confirmed with a negative answer for all τ close to 13 by several authors including Vanden-Broeck (2010) and Champneys et al. (2002). Further, numerical computations of fully nonlinear solutions predicted by the steady KdV equation (9.3.227) for τ < 13 show inaccurate answer in the sense that the solutions do not have flat free profiles in the far field as shown in Figure 9.4, but are generalized solitary waves as in Figure 9.6. The fifth order KdV equation also predicts the generalized solitary waves shown in Figure 9.6 when their amplitude is small and τ is close to 13 .
9.4 Solutions of the KdV Equation: Solitons and Cnoidal Waves To find solutions of the KdV equation, it is convenient to rewrite it in terms of dimensional variables as 3 ch2 ηt + c 1 + η ηx + ηxxx = 0, (9.4.1) 2h 6
9.4 Solutions of the KdV Equation: Solitons and Cnoidal Waves
459
√ where c = gh, and the total depth H = h + η. The first two terms (ηt + cηx ) describe wave evolution at the shallow water speed c, the third term with coefficient (3c/2h) represents a nonlinear wave steepening, and the last term with coefficient (ch2 /6) describes linear dispersion. Thus, the KdV equation is a balance between time evolution, nonlinearity, and linear dispersion. The dimensional velocity u is obtained from (9.3.55) with (9.3.58) in the form c 1 2 h u= (9.4.2) η− η + ηxx . h 4h 3 We seek a traveling wave solution of (9.4.1) in the frame X so that η = η(X) and X = x − U t with η → 0, as |x| → ∞, where U is a constant speed. Substituting this solution in (9.4.1) gives (c − U )η +
3c ch2 ηη + η = 0, 2h 6
(9.4.3)
where η = dη/dX. Integrating this equation with respect to X yields 3c 2 ch2 η + η = A, 4h 6
(c − U )η +
(9.4.4)
where A is an integration constant. We multiply this equation by 2η and integrate again to obtain (c − U )η + 2
2 2 dη c ch 3 η + = 2Aη + B, 2h 6 dX
(9.4.5)
where B is also a constant of integration. We now consider a special case when η and its derivatives tend to zero at infinity and A = B = 0, so that (9.4.5) gives
dη dX
2 =
3 2 η (a − η), h3
where a = 2h
(9.4.6)
U −1 . c
(9.4.7)
The right-hand side of (9.4.6) vanishes at η = 0 and η = a, and the exact solution of (9.4.6) is given by 2
η(X) = a sech (bX), Thus,
η(x, t) = a sech
2
3a 4h3
b= 1/2
3a 4h3
1/2 .
(9.4.8ab)
(x − U t) ,
(9.4.9)
460
9 Solitons and the Inverse Scattering Transform
where the velocity of the wave is a . U =c 1+ 2h
(9.4.10)
This is an exact solution of the KdV equation for all a/h; however, the equation is derived with the approximation (a/h) 1. The solution (9.4.9) is called a soliton (or solitary wave) describing a single hump of height a above the undisturbed depth h and tending rapidly to zero away from X = 0. The solitary wave propagates to the right with velocity U (> c), which is directly proportional to the amplitude a and has width b−1 = (3a/4h3 )−1/2 , that is, b−1 is inversely proportional to the square root of the amplitude a. Another significant feature of the soliton solution is that it travels in the medium without change of shape, which is hardly possible without retaining δ-order terms in the governing equation. A solitary wave profile has already been shown in Figure 9.1. In the general case, when both A and B are nonzero, (9.4.5) can be rewritten as h3 3
dη dX
2
= −η 3 + 2h
U 2h − 1 η2 + (2Aη + B) ≡ F (η), c c
(9.4.11)
where F (η) is a cubic with simple zeros. We seek a real bounded solution for η(X), which has a minimum value zero and a maximum value a and oscillates between the two values. For bounded solutions, all three zeros η1 , η2 , η3 must be real. Without loss of generality, we set η1 = 0 and η2 = a. Hence, the third zero must be negative so that η3 = −(b−a) with b > a > 0. With these choices, F (η) = η(a − η)(η − a + b), and equation (9.4.11) assumes the form 2 h3 dη = η(a − η)(η − a + b), (9.4.12) 3 dX 2a − b , U =c 1+ 2h
where
(9.4.13)
which is obtained by comparing the coefficients of η 2 in (9.4.11) and (9.4.12). Writing a − η = p2 , it follows from equation (9.4.12) that
1/2 dp 3 dX = . (9.4.14) 3 2 4h [(a − p )(b − p2 )]1/2 √ Substituting p = aq in (9.4.14) gives the standard elliptic integral of the first kind (see Dutta and Debnath 1965)
3b 4h3
1/2
q
X= 0
dq , [(1 − q 2 )(1 − m2 q 2 )]1/2
m=
1/2 a , b
and then, function q can be expressed in terms of the Jacobian sn function
(9.4.15)
9.4 Solutions of the KdV Equation: Solitons and Cnoidal Waves
461
Fig. 9.7 A cnoidal wave.
q(X, m) = sn
3b 4h3
1/2
X, m ,
where m is the modulus of sn(z, m). Finally, 1/2 1/2 3b 3b 2 η(X) = a 1 − sn 2 X = acn X , 4h3 4h3
(9.4.16)
(9.4.17)
where cn(z, m) is also the Jacobian elliptic function with a period 2K(m), where K(m) is the complete elliptic integral of the first kind defined by π/2 −1/2 K(m) = 1 − m2 sin2 θ dθ, (9.4.18) 0 2
2
and cn (z) + sn (z) = 1. It is important to note that cnz is periodic, and hence, η(X) represents a train of periodic waves in shallow water. Thus, these waves are called cnoidal waves with wavelength 3 1/2 4h K(m). (9.4.19) λ=2 3b The upshot of this analysis is that solution (9.4.17) represents a nonlinear wave whose shape and wavelength (or period) all depend on the amplitude of the wave. A typical cnoidal wave is shown in Figure 9.7. Sometimes, the cnoidal waves with slowly varying amplitude are observed in rivers. More often, wavetrains behind a weak bore (called an undular bore) can be regarded as cnoidal waves. Two limiting cases are of special physical interest: (i) m → 0 and (ii) m → 1. In the first case, sn z → sin z, cn z → cos z as m → 0 (a → 0). This corresponds to small-amplitude waves where the linearized KdV equation is appropriate. So, in this limiting case, the solution (9.4.17) becomes 1/2 3b 1 , (9.4.20) η(x, t) = a 1 + cos(kx − ωt) , k = 2 h3 where the corresponding dispersion relation is 1 ω = U k = ck 1 − k2 h2 . 6
(9.4.21)
This corresponds to the first two terms of the series expansion of (gk tanh kh)1/2 . Thus, these results are in perfect agreement with the linearized theory.
462
9 Solitons and the Inverse Scattering Transform
In the second limiting case, m → 1(a → b), cnz −→ sech z. Thus, the cnoidal wave solution tends to the classical KdV solitary-wave solution where the wavelength λ, given by (9.4.19), tends to infinity because K(a) = ∞ and K(0) = π/2. The solution identically reduces to (9.4.9) with (9.4.10). We next report the numerical computation of the KdV equation (9.4.1) due to Berezin and Karpman (1966). In terms of new variables defined by x∗ = x − ct,
t∗ = t,
η∗ =
3c η, 2h
(9.4.22)
omitting the asterisks, equation (9.4.1) becomes (9.4.23)
ηt + ηηx + βηxxx = 0,
where β = ( 16 )ch2 . We examine the numerical solution of (9.4.23) with the initial condition x , (9.4.24) η(x, 0) = η0 f where η0 is constant and f (ξ) is a nondimensional function characterizing the initial wave profile. It is convenient to introduce the dimensionless variables ξ=
x ,
τ=
η0 t ,
u=
η η0
(9.4.25)
so that equations (9.4.23) and (9.4.24) reduce to uτ + uuξ + σ −2 uξξξ = 0, u(ξ, 0) = f (ξ),
(9.4.26) (9.4.27) 1
where the dimensionless parameter σ is defined by σ = ( ηβ0 ) 2 . Berezin and Karpman (1966) obtained the numerical solution of (9.4.26) with the Gaussian initial pulse of the form u(ξ, 0) = f (ξ) = exp(−ξ 2 ) and values of the parameter σ = 1.9 and σ = 16.5. Their numerical solutions are shown in Figure 9.8. As shown in Figure 9.8 for case (a), the perturbation splits into a soliton and a wavepacket. In case (b), there are six solitons. It is readily seen that the peaks of the solitons lie nearly on a straight line. This is due to the fact that the velocity of the soliton is proportional to its amplitude, so that the distances traversed by the solitons would also be proportional to their amplitudes. Zabusky’s (1967) numerical investigation of the interaction of two solitons reveals that the taller soliton, initially behind, catches up to the shorter one, they undergo a nonlinear interaction and, then, emerge from the interaction without any change in shape and amplitude. The end result is that the taller soliton reappears in front and the shorter one behind. This is essentially strong computational evidence of the stability of solitons.
9.4 Solutions of the KdV Equation: Solitons and Cnoidal Waves
463
Fig. 9.8 The solutions of the KdV equation u(x, t) for large values of t with the values of the similarity parameter σ: (a) σ = 1.9, (b) σ = 16.5. From Berezin and Karpman (1966).
Using the transformation x∗ = εβ(x − ct),
t∗ = ε3 t,
−1 η ∗ = αε2 η
with αβ(= 3c/2h) = 6 and β 3 (= ch2 /6) = 1, we write the KdV equation (9.4.1) in the normalized form, dropping the asterisks, ηt + 6ηηx + ηxxx = 0.
(9.4.28)
We next seek a steady progressive wave solution of (9.4.28) in the form η = 2k2 f (X),
X = kx − ωt.
(9.4.29)
Then, the equation for f (X) can be integrated once to obtain f (X) = 6A + (4 − 6B)f − 6f 2 ,
(9.4.30)
where A is a constant of integration and the frequency ω is given by ω = k 3 (4 − 6B).
(9.4.31)
A single-soliton solution corresponds to the special case A = B = 0 and is given by f (X) = sech2 (X − X0 ),
(9.4.32)
where X0 is a constant. Thus, for a series of solitons spaced 2σ apart, we write f (X) =
∞
sech2 (X − 2nσ).
(9.4.33)
n=−∞
This is a 2σ periodic function that satisfies (9.4.30) for some A and B. The general elliptic function solution of (9.4.28) can be obtained from the integral of (9.4.30), which can be written as
464
9 Solitons and the Inverse Scattering Transform 2 fX = −4C + 12Af + (4 − 6B)f 2 − 4f 3 ,
(9.4.34)
where C is a constant of integration. Various asymptotic and numerical results lead to the relations (see Whitham 1984) C=−
1 d2 B(σ) 1 dA(σ) , = 2 dσ 4 dσ 2
(9.4.35ab)
and the cubic in (9.4.34) can be factorized as 3 −C + 3Af + 1 − B f 2 − f 3 = (f1 − f )(f − f2 )(f − f3 ), 2
(9.4.36)
where fr (σ) (r = 1, 2, 3) are determined from A(σ), B(σ), and C(σ). If we set f1 > f2 > f3 and, then, the periodic solution oscillates between f1 at X = 0 and f2 at X = σ, one particular form of the solution is given by (f1 − f3 )X , (9.4.37) f (X) = f2 + (f1 − f2 )cn2 where the modulus m of cn(z, m) is given by f1 − f2 m2 = . f1 − f3
(9.4.38)
Thus, it follows from (9.4.33) and (9.4.37) that the following identity holds: ∞ f2 + (f1 − f2 )cn (f1 − f3 )X = sech2 (X − 2nσ), 2
(9.4.39)
n=−∞
which can be verified by comparing the periods and poles of the two sides. It is interesting to point out that there is also a theta function representation of the periodic solution of the KdV equation. The solution may be written in the form f (X) =
∂2 log Θ(X), ∂X 2
(9.4.40)
where Θ(X) is a slight modification of the theta function ϑ4 (z, q) in Dutta and Debnath’s (1965) notation, the imaginary period τ in q = exp(iπτ ) has been set equal to (2iσ/π) so that q = exp(−2σ), and the independent variable z is changed to (iX + π/2) and ∞ exp −2nX − 2n2 σ . (9.4.41) Θ(X) = n=−∞
A simple direct proof of (9.4.40) is not obvious. However, it is equivalent to (9.4.33), which can be verified by using Jacobi’s infinite product formula for the theta functions (Dutta and Debnath 1965, p.108). With the appropriate modification in (9.4.29), the product representation of Θ(X) is
9.4 Solutions of the KdV Equation: Solitons and Cnoidal Waves
Θ(X) = G
∞
2X−(2r−1)2σ
1+e
∞
r=1
1 + e−2X−(r−1)2σ .
465
(9.4.42)
r=1
Thus, the series representation (9.4.41) follows from the preceding analysis by replacing X by X + σ. There is a similar situation for the modified KdV equation in the normalized form vr + 3v 2 vx + vxxx = 0. If we seek a steady progressive solution in the form √ v = 2kf (X), X = kx − ωt,
(9.4.43)
(9.4.44)
then f (X) satisfies the equation FXX = 2A + (2B + 1)f − 2f 3 , where A is a constant of integration and ω 2B = − 1 . k3
(9.4.45)
(9.4.46)
A single soliton corresponding to the special case A = B = 0 is given by f (X) = sech(X − X0 ),
(9.4.47)
and hence, a series of solitons spaced 2σ apart is given by f (X) =
∞
exp(X − 2nσ).
(9.4.48)
n=−∞
This is, indeed, an exact solution. The required identity between the series of the sech function is found by substituting (9.4.48) into (9.4.45). It follows from Whitham’s argument (1974) that A = 0,
B=6
∞ cosh(2sσ) . 2 sinh (2sσ) s=1
(9.4.49)
Integrating (9.4.45) once gives 2 fX = −C + (2B + 1)f 2 − f 4 ≡ f12 − f 2 f 2 − f22 ,
(9.4.50)
where C is a constant of integration and f1 ≥ f ≥ f2 . This equation admits the periodic solution (9.4.51) f (X) = f1 dn(f1 X, m), where the modulus m of the elliptic function is m2 = f12 − f22 /f12 ,
(9.4.52)
466
9 Solitons and the Inverse Scattering Transform
and (9.4.48) gives an identity for the dn function in terms of a sum of the sech functions. The relation between σ and m is given by σf1 (σ) = K(m). Finally, the higher-order modified KdV equation vt + (p + 1)v p vx + vxxxx = 0,
p > 2,
(9.4.53)
admits single-soliton solutions in the form v(x, t) = a sech2/p (kx − ωt).
(9.4.54)
However, in view of the fractional powers of the sech function, it seems, perhaps, unlikely that there will be any simple superposition formula.
9.5 The Lie Group Method and Similarity Analysis of the KdV Equation The KdV equation describes the generation and propagation of moderately smallamplitude shallow water waves and many other important phenomena, where a small nonlinearity is combined with a cubic dispersion relation. In many physical situations, its similarity solutions become important. A simple transformation x −→ cx,
t −→ bt,
and
1 + u −→ au,
(9.5.1)
can be used to transform the KdV equation ut + ux + uux + uxxx = 0 into the general form
ut +
ab b uux + 3 uxxx = 0, c c
(9.5.2)
(9.5.3)
where a, b, c are nonzero real constants. With suitable choices of the constants, this equation reduces to the standard form (9.4.28), which is invariant under the transformation Ta defined by x ˜ = ax,
t˜ = a3 t,
u ˜ = a−2 u
(9.5.4)
for a nonzero real a. The set of all such transformations {Ta } forms an infinite Lie group with parameter a with the composition (multiplicative) law Ta Tb = Tab . This law is commutative. Also, the associative law is satisfied since Ta (Tb Tc ) = Ta Tbc = Tabc = Tab Tc = (Ta Tb )Tc . In view of the fact that
T1 Ta = T1a = Ta1 = Ta
for all a = 0, T1 represents the identity transformation.
9.5 The Lie Group Method and Similarity Analysis of the KdV Equation
467
Finally, we find that Ta Ta−1 = T1 = Ta−1 Ta . This confirms that Ta−1 is both a left-hand and right-hand inverse of Ta . The fact that the KdV equation (9.4.28) with 6 replaced by −6 is invariant under a continuous (or Lie) group of transformations Ta suggests that we can seek invariant properties of the solutions. In fact, the quantities u(x, t) = tp f (ξ),
ξ = xtq
(9.5.5ab)
are invariant under the transformation Ta provided p = − 23 and q = − 13 . It is also easy to check that, if u(x, t) = −(3t)2/3 f (ξ) where ξ = x(3t)−1/3 , then f (ξ) satisfies the ordinary differential equation f + (6f − ξ)f − 2f = 0.
(9.5.6)
Another substitution of f = λ(dV /dξ) − V 2 in (9.5.6), where V = V (ξ) and λ is a constant to be determined, yields a Painlevé equation of the second kind V − ξV − 2V 3 = 0.
(9.5.7)
This equation gives a solution describing a wave profile that decays as ξ → +∞ and oscillates as ξ → −∞. The presence of a Painlevé equation, for which each movable singularity is a pole, is not accidental. It is believed that there is a direct correspondence between the appearance of a Painlevé equation for a given partial differential equation and the existence of an inverse scattering transform (and hence, soliton solutions) for that equation. Zakharov and Shabat (1974) proved an exact reduction of the Boussinesq equation 1 1 (9.5.8) utt − uxx − u2 xx − uxxxx = 0 2 4 to an ordinary differential equation by looking for a traveling wave solution u(x, t) = f (x − U t) = f (X),
(9.5.9)
1 1 1 − U 2 f + f 2 + f = 0. 2 4
(9.5.10)
where f (X) satisfies
This can be integrated twice to obtain the following Painlevé equations, depending on the constants of integration: f + 2f 2 + A = 0,
f + 2f 2 + X = 0.
(9.5.11ab)
In general, certain partial differential equations admit similarity solutions in the form u(x, t) = t−α f xt−β , (9.5.12)
468
9 Solitons and the Inverse Scattering Transform
where α and β are constants and f satisfies an ordinary differential equation. The equation is invariant under the transformation Ta : x ˜ = aβ x,
u ˜ = a−α u,
t˜ = at,
(9.5.13)
where a is a scalar. The set of all such transformations Ta also forms a continuous group. It can easily be verified that the normalized KdV equation ut + 6uux + uxxx = 0
(9.5.14)
has the similarity solution u(x, t) = (3t)−2/3 f (ξ),
ξ = x(3t)−1/3 .
(9.5.15ab)
Substituting (9.5.15ab) in (9.5.14) gives the following ordinary differential equation for f (ξ) in the form f + (6f − ξ)f − 2f = 0.
(9.5.16)
In the limit as ξ → ∞, solutions of equation (9.5.16) tend to solutions of the linearized equation (9.5.17) f − 2f − ξf = 0. This equation admits a solution f (ξ) = Ai (ξ), where Ai (ξ) is the Airy function. The other two solutions Bi (ξ), which is exponentially large, and Gi (ξ) ∼ −(πz 2 )−1 do not decay exponentially (Abramowitz and Stegun 1972, pp. 446–450). Thus we set the boundary condition f (ξ) ∼ aAi (ξ),
as ξ → ∞,
(9.5.18)
where a is an amplitude parameter. The numerical computation of Berezin and Karpman (1966) suggests that, when a is small enough, f (ξ) is oscillatory, as ξ → −∞, but otherwise, f (ξ) may develop singularities. Rosales (1978) showed that there exists a critical value a1 of a which separates the oscillatory nature from the singular solutions. For |a| = a1 , f (ξ) ∼ 12 ξ as ξ → −∞. For |a| = a1 , f (ξ) becomes oscillatory, as ξ → −∞. On the other hand, when a > a1 , f (ξ) develops a singularity. Some special nonlinear evolution equations admit solutions that are rational functions of the independent variables. For example, u(x, t) = (x/t) is a rational solution of the equation (9.5.19) ut + uux = 0. It turns out that the KdV equation (9.4.28) with the coefficient 6 replaced by −6 also has a simple rational solution. Assuming that u = u(x) and u, u , u → 0, as |x| → ∞, the resulting KdV equation u − 6uu = 0 can be integrated twice to obtain u 2 = 2u3 , which admits the rational solution u(x, t) = 2x−2 .
(9.5.20)
9.6 Conservation Laws and Nonlinear Transformations
469
This is singular at x = 0. It can also be shown that equation (9.4.28) with the coefficient 6 replaced by −6 has a rational solution in the form u(x, t) =
6x(x3 − 24t) . (x3 + 12t2 )
(9.5.21)
This is also singular on x3 + 12t2 = 0. It is conjectured that all rational solutions of the KdV equation are singular! Zabusky (1967) showed that the modified KdV equation ut + 6u2 ux + uxxx = 0
(9.5.22)
2 u(x, t) = A − (4A)/ 4A2 x − 6A2 t + 1 ,
(9.5.23)
has the rational solution
where A is a real constant.
9.6 Conservation Laws and Nonlinear Transformations An equation of the form ∂T ∂X + =0 (9.6.1) ∂t ∂x is called a conservation law, where T and X are known as the density and the flux, respectively, and neither T nor X involve derivatives with respect to t. If both T and X are integrable on (−∞, ∞), so that X tends to a constant as |x| → ∞, then equation (9.6.1) can be integrated to obtain ∞ d T dx = 0, or equivalently, (9.6.2a) dt −∞ ∞ T dx = const. (9.6.2b) −∞
The integral (9.6.2b) is usually called a constant of motion, provided t is interpreted as a time-like variable. It is convenient to write the KdV equation in the standard form that is much used for the development of the theory of solitons, Ku ≡ ut − 6uux + uxxx = 0.
(9.6.3)
This is already in conservation form with T =u
and X = uxx − 3u2 .
Hence, if T and X are integrable, and u satisfies (9.6.3), then
(9.6.4ab)
470
9 Solitons and the Inverse Scattering Transform
∞
u dx = const.,
(9.6.5)
−∞
which applies to all solutions provided that u and its gradients tend to zero as |x| → ∞. Thus, the first constant of motion is simply the spatial integral of u. The second conservation law for (9.6.3) can be obtained by multiplying it by u, so that 1 2 ∂ 1 2 ∂ 3 (9.6.6) u + uuxx − ux − 2u = 0. ∂t 2 ∂x 2 Thus,
∞
−∞
1 2 u dx = const. 2
(9.6.7)
for all solutions of the KdV equation which vanish fast enough at infinity. Thus, both u and u2 are conserved densities for the motion associated with the KdV equation. We know that the KdV equation describes a certain class of nonlinear water waves. In fact, u is associated with the free surface elevation function which, in turn, is also proportional to the velocity in the x-direction. Thus, equation (9.6.5) describes the conservation of mass, and (9.6.7) the conservation of horizontal momentum for water waves within the scope of a shallow water approximation theory. This immediately suggests that there should be a corresponding conserved density that could be associated with the energy of water waves. Indeed, this is true and can be confirmed by adding 3u2 × (9.6.3) to ux × {(∂/∂x)(9.6.3)} so that the resulting equation can be rewritten as 9 4 ∂ ∂ 1 2 1 2 3 2 2 u + ux + − u + 3u uxx − 6uux + ux uxxx − uxx = 0. (9.6.8) ∂t 2 ∂x 2 2 This gives a third constant of motion, ∞ 1 2 3 u + ux dx = const. 2 −∞
(9.6.9)
Then, by using very laborious methods, Miura et al. (1968) obtained eight more conservation laws for the KdV equation. They then conjectured that there exist an infinite number of polynomial conservation laws for the KdV equation. At the same time, a question was raised: Is the existence of soliton solutions of the KdV equation closely related to the existence of an infinite number of conservation laws? Miura et al. (1968) developed an ingenious method of determining an infinite number of conservation laws by introducing the Miura transformation, u = v 2 + vx .
(9.6.10)
This is similar to the Cole–Hopf transformation used in the Burgers equation which can be reduced to the linear diffusion equation. A direct substitution of (9.6.10) in (9.6.3) gives 2vvt + vxt − 6 v 2 + vx (2vvx + vxx ) + 6vx vxx + 2vvxxx + vxxxx = 0,
9.6 Conservation Laws and Nonlinear Transformations
471
which can then be put into the form ∂ M v = 0, 2v + ∂x where the operator M is defined by M v ≡ vt − 6v 2 vx + vxxx = 0.
(9.6.11)
This is called the modified KdV (mKdV) equation. Thus, if M v = 0, then u satisfies the KdV equation (9.6.3). In contrast to the Cole–Hopf transformation, the Miura transformation reduces one nonlinear equation to another nonlinear equation, neither of which can easily be solved. However, the Miura transformation establishes a connection between the KdV equation and the Sturm–Liouville problem, and hence, it leads to the inverse scattering method for the exact solution of the initial-value problem for the KdV equation. Another remarkable fact about the Miura transformation is that there exists an infinite number of conservation laws. It is now convenient to work with w, rather than v, where v = 12 ε−1 + εw and ε is an arbitrary real parameter. Consequently, the Miura transformation becomes u=
1 −2 ε + w + εwx + ε2 w2 . 4
(9.6.12)
Since any arbitrary constant can be incorporated in the solution for u, we just write u = w + εwx + ε2 w2 . (9.6.13) This is known as the Gardner transformation. It is important to note that, since v exists, so does w. Substitution of (9.6.13) in (9.6.3) gives 0 = ut − 6uux + uxxx = wt + εwxt + 2ε2 wwt − 6 w + εwx + ε2 w2 wx + εwxx + 2ε2 wwx + wxxx + εwxxxx + 2ε2 (wwx )xx
∂ + 2ε2 w wt − 6 w + ε2 w2 wx + wxxx . = 1+ε ∂x Thus, w satisfies the Gardner equation wt − 6 w + ε2 w2 wx + wxxx = 0
(9.6.14)
for all ε. We note that, when ε = 0, this equation becomes the KdV equation with w = u. Another interesting fact is that equation (9.6.14) is already in the conservation form ∂ ∂w (9.6.15) + wxx − 3w 2 − 2ε2 w3 = 0, ∂t ∂x
472
9 Solitons and the Inverse Scattering Transform
with w as the conserved density, that is, ∞ w dx = const.
(9.6.16)
−∞
To obtain an infinite number of conservation laws for the KdV equation, we use the arbitrary parameter ε. Since w → u as ε → 0, we write an asymptotic expansion of w(x, t, ε) in ε: ∞ εn wn (x, t) as ε → 0. (9.6.17) w∼ n=0
Treating the constant in (9.6.16) similarly, as a power series in ε, and using (9.6.17), we obtain an infinite number of constants of motion. Finally, we substitute (9.6.17) in the Gardner transformation (9.6.13) and, then, equate coefficients of εn , for each n = 0, 1, 2, . . . . It is then easy to see that w1 = −w0x = −ux , w2 = −w1x = −w02 = uxx − u2 , w0 = u, 2 w3 = −w2x − 2w0 w1 = − uxx − u x + 2uux ,
w4 = w3x − 2w0 w1 − w12 = − 2uux − uxx − u2 x x − 2u uxx − u2 − u2x , and so on. In particular, the integrals which are generated by w0 , w2 , and w4 become the first three integrals (9.6.5), (9.6.7), and (9.6.9), respectively. On the other hand, w1 and w3 are exact differentials in x, and hence, the corresponding integrals ∞ wn dx = const. (9.6.18) −∞
will not give us any useful result. In other words, if n is even, then (9.6.18) generates an infinite set of conservation laws, and, if n is odd, wn is an exact differential and, hence, does not lead to any conservation law. Much of the significant work on the theory of the KdV equation was initiated by the publication of several papers of Gardner et al. (1967, 1974). We close this section by including a few conservation laws for the Boussinesq equation, the KP equation, and shallow water equations. It is convenient to introduce ηt = −ux in the Boussinesq equation (9.3.202) so that it becomes (9.6.19) −utx − ηxx + 3 η 2 xx − ηxxxx = 0. Integrating this equation with respect to x and invoking the decay conditions as |x| → ∞, the Boussinesq equation (9.3.202) can be rewritten as the pair of equations ηt = −ux , ut + ηx − 3η 2 x + ηxxx = 0. (9.6.20) Consequently, it turns out that ∞ ηt dx = −[u]∞ −∞ , −∞
∞
−∞
∞ ut dx = 3η 2 − η − ηxx −∞ .
9.6 Conservation Laws and Nonlinear Transformations
Thus, we obtain the conservation of mass in the form ∞ η dx = const.,
473
(9.6.21)
−∞
and the conservation of momentum as ∞ ut dx = const.
(9.6.22)
−∞
We next recall the KP equation (9.3.135) in the form 6ut + 9uux + uxx + 3vy = 0,
vx = uy .
(9.6.23)
Integrating the second equation with respect to x and invoking the decay conditions at infinity gives ∞ ∞ ∂ u dx = 0, and hence, u dx = c(t). (9.6.24) ∂y −∞ −∞ Physically, the N -soliton solution of the KP equation describes the interaction of waves so that the function c(t) represents a constant and (9.6.24) is also a constant. Consequently, ∞ u dx = const. (9.6.25) −∞
A similar argument leads to the result ∞ v dy = const.
(9.6.26)
−∞
The above results are similar to (9.6.21), (9.6.22). Integrating the first equation in (9.6.23) yields ∞ ∞ ∞ ∂ ∂ u dx + uxx − 3u2 −∞ + 3 v dx = 0. ∂t −∞ ∂y −∞
(9.6.27)
Using (9.6.25) and suitable decay conditions at infinity gives another conserva ∞ tion law v dx = const. (9.6.28) −∞
Clearly, (9.6.26)–(9.6.28) represent momentum conservation laws. Finally, we derive the conservation laws for shallow water waves. We recall shallow water equations (2.7.67)–(2.7.70) in (1 + 1) dimensions with δ → 0 and ε = 1 so that 1 + εη = h(x, t). Consequently, these equations become (9.6.29)
ut + uux + wuz + hx = 0, ux + wz = 0, w = ht + uhx
on z = h,
and w = 0
Integrating (9.6.30) with respect to z from 0 to h gives
on z = 0.
(9.6.30) (9.6.31)
474
9 Solitons and the Inverse Scattering Transform
h
ux dz + [w]h0 = 0
∂ ∂x
h
u dz + ht = 0.
(9.6.32)
0
Making use of suitable decay conditions, we obtain the law of mass conservation as ∞ h(x, t) dx = const. (9.6.33) −∞
We next multiply (9.6.30) by u and add the result to (9.6.29) to find ut + 2uux + (uw)z + hx = 0. Integrating this equation with respect to z from 0 to h so that h h ∂ 1 2 ∂ 2 u dz + u dz = 0, h + ∂t 0 ∂x 2 0 leads to the law of conservation of momentum as ∞ h u dz dx = const. −∞
(9.6.34)
(9.6.35)
(9.6.36)
0
Multiplying (9.6.29) by u, we obtain 1 3 1 2 1 1 2 + + u u u w − u2 wz + (uh)x − hux = 0. 2 3 2 2 t x z
(9.6.37)
We substitute wz and ux from (9.6.30) and use the fact that h = h(x, t) to obtain 1 2 1 3 1 2 u wu + hw + uh + u + = 0. (9.6.38) 2 2 2 t x z Integrating this result with respect to z and interchanging differentiation under the integral sign gives h h 1 2 1 3 2 h + u dz + (9.6.39) uh + u dz = 0. 2 3 0 0 t x This result is in the conservation form and yields the law of conservation of energy h ∞ u2 dz dx = const. (9.6.40) h2 + −∞
0
9.7 The Inverse Scattering Transform (IST) Method Historically, Gardner et al. (1974) formulated an ingenious method for finding the exact solution of the KdV equation. This novel method has been generalized to solve
9.7 The Inverse Scattering Transform (IST) Method
475
several other nonlinear equations. In this section, we describe the inverse scattering transform for the KdV equation. The exact solution of the KdV equation is obtained by associating its solution with the potential of a time-independent Schrödinger equation. The next step is the solution of the quantum mechanical problem with the initial value for the KdV equation taken as the potential. This involves calculations of the discrete (bound) eigenfunctions, their normalization constants and eigenvalues, and the reflection and transmission coefficients of the continuous (unbounded) states. These results are collectively known as the scattering data of the Schrödinger equation. The next step is to determine the evolution of the scattering data for any potential that evolves, according to the KdV equation, from a prescribed initial function. Both discrete and continuous eigenvalues are found to be invariant under such changes, and the normalization constants and the reflection and transmission coefficients evolve according to simple exponential laws. The final step of the method deals with the determination of the potential for any time t from the inversion of the scattering data. The main problem is to find the exact solution of the general initial-value problem for the KdV equation ut − 6uux + uxxx = 0,
x ∈ R, t > 0,
(9.7.1)
x ∈ R,
(9.7.2)
with the initial condition u(x, 0) = u0 (x),
where u0 (x) satisfies certain fairly weak conditions so that the solution u(x, t) exists for all x and t. We begin with the Miura (1968) transformation defined by u = v 2 + vx ,
(9.7.3)
where v = v(x, t). Substituting (9.7.3) in (9.7.1) gives 2vvt + vxt − 6 v 2 + vx (2vvx + vxx ) + 6vx vxx + 2vvxxx + vxxxx = 0. This can be rearranged to obtain equation ∂ + 2v vt − 6v 2 vx + vxxx = 0. ∂x
(9.7.4)
Thus, if v is a solution of the modified KdV equation, then (9.7.3) defines a solution of the KdV equation. We note here that, if u is given, equation (9.7.3) represents the Riccati equation for v, so that the transformation v=
ψx , ψ
(9.7.5)
where ψ(x, t) is a nonzero differentiable function, reduces (9.7.3) to the form
476
9 Solitons and the Inverse Scattering Transform
ψxx − uψ = 0.
(9.7.6)
The function u involved in (9.7.6) is a solution of the KdV equation (9.7.1). We further note that KdV equation (9.7.1) is Galilean invariant, that is, it is invariant under the transformation x ˜ = x + 6λt, u ˜ = u + λ, where λ ∈ R, so that we can replace u by u − λ to obtain a second-order linear equation Lψ = λψ,
(9.7.7)
where the operator L is given by L≡−
∂ + u. ∂x2
(9.7.8)
This is the Schrödinger equation for ψ with the potential u(x, t) and eigenvalue λ = λ(t) which depends on t because of the parametric dependence on t. However, this Schrödinger equation for ψ is different from the one in quantum mechanics because the potential u(x, t) is not known and it is the solution of the KdV equation. So, the method of solution (9.7.7) is not so simple. However, it is well known that equation (9.7.7) admits two distinct kinds of solutions recognized as bound states, where ψ → 0 exponentially as |x| → ∞, and scattering states, where ψ oscillates with x at infinity. Since both u and λ depend on t, the eigenfunctions ψ(x, t) also depend on t. The quantity t involved in (9.7.7) cannot be regarded as time, but simply as a parameter, so that u = u(x, t) represents a family of potentials. At t = 0, the initial condition is given by u(x, 0) = u0 (x), and hence, the problem is first to solve the one-dimensional Schrödinger equation for the known potential u0 (x), 2 ∂ − u0 (x) ψ(x, t) = λ(t)ψ(x, t). (9.7.9) − ∂x2 This will determine how the eigenfunctions and eigenvalues evolve as potential changes from the given data according to the KdV equation (9.7.1). The potential u(x, t) for any t > 0 can then be determined from the scattering data at time t by the inverse scattering transformation. To determine the eigenvalues λ(t), we first differentiate (9.7.7) with respect to x to obtain ψxxx − ux ψ − (u − λ)ψx = 0, and then with respect to t to derive ψxxt − (ut − λt )ψ − (u − λ)ψt = 0.
(9.7.10)
The quantity ut can be eliminated from (9.7.10) by means of (9.7.1) so that (9.7.9) becomes (−L + λ)ψt + (uxxx − 6uux )ψ + λt ψ = 0, (9.7.11) where the term (uxxx ψ) can be obtained from the result (ux ψ)xx = uxxx ψ + ux ψxx + 2uxx ψx .
(9.7.12)
9.7 The Inverse Scattering Transform (IST) Method
477
We solve this equation for (uxxx ψ) so that the resulting expression can be expressed by the original Schrödinger equation in the form uxxx ψ = (ux ψ)xx − ux ψxx − 2uxx ψx = (−L + λ)(ux ψ) − 2uxx ψx .
(9.7.13)
Substituting this result in (9.7.11) yields (−L + λ)(ψt + ux ψ) − 2(3uux ψ + uxx ψx ) + λt ψ = 0.
(9.7.14)
To replace the quantity (3uux ψ + uxx ψx ), we rewrite the term uxx ψx by first using the identity (9.7.15) (uψx )xx = uxx ψx + 2ux ψxx + uψxxx . The quantity ψxx in the second term on the right-hand side of (9.7.15) can be eliminated by using the Schrödinger equation, and ψxxx in the third term can be eliminated by the x-derivative of the Schrödinger equation, so that (9.7.15) becomes (uψx )xx = uxx ψx + 2ux (u − λ)ψ + u ux ψ + (u − λ)ψx (9.7.16) = uxx ψx − 2λux ψ + 3uux ψ + (u − λ)uψx . We next solve for (3uux ψ + uxx ψx ) by means of the Schrödinger equation and its x-derivatives, so that 3uux ψ + uxx ψx = 3uux ψ + (uψx )xx + 2λux ψ − 3uux ψ − (u − λ)uψx = (−L + λ)(uψx ) + 2λux ψ = (−L + λ)(uψx ) + 2λ ψxxx − (u − λ)ψx = (−L + λ)(uψx + 2λψx ). Substituting (9.7.17) in (9.7.14) gives (−L + λ) ψt + ux ψ − 2(u + 2λ)ψx + λt ψ = 0.
(9.7.17)
(9.7.18)
We next introduce a new quantity Ψ defined by Ψ = ψt + ux ψ − 2(u + 2λ)ψx ,
(9.7.19)
which, by replacing (ux ψ) with the derivative of (9.7.7) with respect to x, is = ψt + ψxxx − 3(u + λ)ψx .
(9.7.20)
In view of (9.7.19), equation (9.7.18) assumes the form (−L + λ)Ψ = −λt ψ.
(9.7.21)
Equations (9.7.19)–(9.7.21) are the main results for determining the behavior of all quantities involved in the scattering problem as the potential evolves according to the KdV equation.
478
9 Solitons and the Inverse Scattering Transform
As stated earlier, the solution of the inverse scattering problem requires a priori knowledge of the scattering data. This includes the discrete eigenvalues, the normalized coefficients for the eigenfunctions, and the reflection and transmission coefficients. We first consider the discrete spectrum of eigenvalues (bound states). In view of the fact that u(x, t) decays rapidly as |x| → ∞ for all t, the Schrödinger equation (9.7.7) admits a finite number of discrete eigenstates with negative energy λn = −κ2n , n = 1, 2, . . . , N , and a continuous spectrum of positive energy (λ = κ2 > 0). The corresponding discrete eigenfunctions ψn belong to the Hilbert space L2 (R). To determine the evolution of the eigenvalues, equation (9.7.21) is multiplied by ψn , and we use (9.7.7) to simplify the result to obtain −λnt ψn2 (x) = ψn (x)(−L + λn )Ψ = ψn Ψxx − Ψ ψnxx ∂Ψ ∂ψn ∂ ψn . −Ψ = ∂x ∂x ∂x Integrating this equation with respect to x over (−∞, ∞) gives ∞ ∞ ∂Ψ ∂ψn 2 −Ψ ψn (x) dx = ψn . −λnt ∂x ∂x −∞ −∞ Normalizing the eigenfunctions ψn (x) of the bound states by ∞ 12 2 ψn (x) dx = 1 ψn =
(9.7.22)
(9.7.23)
(9.7.24)
−∞
and using the fact that ψn (x) and its spatial derivatives vanish as x → ±∞, we conclude that the right-hand side of (9.7.23) vanishes so that dλn = 0, dt
(9.7.25)
that is, for a potential u(x, t) that evolves with t according to the KdV equation, the eigenvalues of the bound states are invariant with respect to t : λn (t) = λn (0) for n = 1, 2, . . . , N . The invariance of the discrete eigenvalues represented by (9.7.25) has several important consequences. First, equation (9.7.21) with λnt (t) = 0 implies that Ψn represents the discrete eigenfunctions of the Schrödinger equation (9.7.21) with the eigenvalues λ = λn . Second, it follows that (9.7.21) combined with (9.7.19) leads to the equation (9.7.26) ψnt + ux ψn − 2 u − 2κ2n ψnx = 0. This can be regarded as a time evolution equation for ψn (x, t). The bounded solution for the nth eigenfunction can then be represented from its asymptotic behavior in the form ψn (x, t) ∼ cn (t) exp(−κn x), as x → +∞, (9.7.27) where the real constants cn are determined by the normalized condition (9.7.24). Substituting this asymptotic form (9.7.27) in (9.7.26) and imposing a major restriction u(x, t) → 0 as |x| → ∞ leads to the equation for cn (t) in the form
9.7 The Inverse Scattering Transform (IST) Method
479
Fig. 9.9 Sketch representing the scattering by a potential u.
d cn (t) − 4κ3n cn (t) = 0. dt
(9.7.28)
This equation admits solutions given by cn (t) = cn (0) exp 4tκ3n ,
(9.7.29)
where cn (0), n = 1, 2, . . . , N are the normalization constants determined at t = 0. Thus, the asymptotic representation of ψn (x, t) is explicitly given by (9.7.27) with (9.7.29). As a passing remark, we note that the imposed restriction on u(x, t) limits the theory of the KdV soliton and eliminates solutions corresponding to other nonlinear waves. For a continuous spectrum of eigenvalues (unbounded states), the eigenvalue λ = k 2 > 0, where k may take any real value. We visualize the wave function ψ as the spatially dependent part of a steady plane wave impinging on the potential u from x → +∞ for the time-dependent Schrödinger equation. Physically, there will be an interaction of the plane wave with the potential allowing for the existence of scattering states that can be treated as the linear combination of a transmitted wave exp(−ikx) at x = −∞ of amplitude a(k) and a reflected wave exp(ikx) at x = +∞ of amplitude b(k). In other words, the wave function for the continuous spectrum has asymptotic behavior in the form ψ(x) ∼ exp(−ikx) + b(k) exp(ikx) ψ(x) ∼ a(k) exp(−ikx)
as x → ∞,
as x → −∞,
(9.7.30a) (9.7.30b)
where a(k) and b(k) are called the transmission and reflection coefficients, respectively, and can be determined uniquely from the initial data. The term exp(−ikx) of unit amplitude in (9.7.30a) is a traveling wave from x = +∞. In general, both a(k) and b(k) are complex functions of a real variable k. A sketch representing the scattering by a potential u is shown in Figure 9.9. We next consider two distinct eigenfunctions for the same potential u so that − κ2m + u ψm = 0 and ψn − κ2n + u ψn = 0, (9.7.31ab) ψm where the primes denote derivatives with respect to x.
480
9 Solitons and the Inverse Scattering Transform
Clearly,
d W (ψn , ψm ), κ2m − κ2n ψm ψn = ψn ψm − ψm ψn = dx
where W is the Wronskian of ψn and ψm . Integrating (9.7.32) gives, for m = n, 2 ∞ ∞ 2 κm − κn ψm ψn dx = W (ψn , ψm ) −∞ = 0,
(9.7.32)
(9.7.33)
−∞
because ψn , ψm → 0 as |x| → ∞. This shows that ψm , ψn are orthogonal. The continuous eigenfunction ψ is also orthogonal to each discrete eigenfunction ψm . Thus, the discrete and continuous eigenfunctions together form a complete orthogonal set, and hence, any function ψ ∈ L2 (R) can be represented as a linear combination of all the ψm ’s plus an integral of ψ(x, k) over all k. If φ and ψ are two solutions of (9.7.7) with the same value of λ = k 2 (> 0), then equation (9.7.7) implies that the derivative of W is zero, that is, W (φ, ψ) is constant. With an additional restriction that φ is proportional to ψ, it turns out that W (φ, ψ) = 0 for all x. Denoting ψ ∗ as the complex conjugate of the continuous eigenfunction ψ, we can compute W (ψ, ψ ∗ ) at both x = ±∞ and, then, use (9.7.30a) and (9.7.30b) to obtain (9.7.34) W ψ, ψ ∗ = 2ikaa∗ = 2ik 1 − bb∗ . Or equivalently, |a|2 + |b|2 = 1.
(9.7.35)
This is a statement of the conservation of energy in the theory of scattering. A procedure similar to the case of the discrete spectrum can be employed to determine the scattering coefficients. Integrating (9.7.21) once with respect to x gives Ψ ψx − ψΨx = g(t; k),
(9.7.36)
where Ψ is defined by (9.7.19) and g(t; k) is an arbitrary function resulting from integration. For the continuous eigenfunction, ψ(x, k; t) is given by (9.7.30b), and hence, the asymptotic result for Ψ follows from (9.7.19): Ψ ∼ at + 4iak 3 e−ikx as x → −∞. (9.7.37) Consequently, (Ψ ψx − ψΨx ) → 0
as x → −∞.
(9.7.38)
Obviously, g(t; k) = 0 for all time t, and (9.7.36) can be integrated to obtain (Ψ/ψ) = h(t; k),
or equivalently,
Ψ = ψh,
(9.7.39)
where h is another arbitrary function of integration. The use of asymptotic results for ψ and Ψ as x → −∞ in (9.7.39) yields the equation
9.7 The Inverse Scattering Transform (IST) Method
da + 4iak3 = ha. dt
481
(9.7.40)
Similarly, we next use (9.7.30a) and the asymptotic form of Ψ , as x → −∞, in the form db (9.7.41) Ψ ∼ eikx + 4ik3 eikx − be−ikx dt in (9.7.39) to obtain db ikx e + 4ik3 e−ikx − beikx = h e−ikx + beikx . dt
(9.7.42)
Thus, it follows from the linear independence of eikx and e−ikx that db − 4ik3 b = hb dt
and h = 4ik3 ,
(9.7.43ab)
and therefore, equation (9.7.40) gives da = 0. dt
(9.7.44)
Finally, the solutions for a and b are given by a(k; t) = a(k; 0)
and
b(k; t) = b(k; 0) exp 8ik3 t ,
t > 0.
(9.7.45)
This completes the determination of the scattering data S(t), which are summarized as follows:
cn (t) = cn (0) exp(4κ3n t), κn = const.; (9.7.46) [1ex]a(k; t) = a(k; 0); b(k; t) = b(k; 0) exp(8ik3 t), where cn (0), a(k; 0), and b(k; 0) are determined from the initial condition for the KdV equation. The simplest approach to solving the inverse scattering problem involves the integral representation of solutions of the Schrödinger equation (9.7.7) with λ = k2 , Lψ = k 2 ψ.
(9.7.47)
It is convenient to introduce the following integral representation of the solution of (9.7.47), ∞ K(x, z) exp(ikz) dz, (9.7.48) φk (x) = exp(ikx) + x ∞ K(x, z) exp(−ikz) dz. (9.7.49) φ−k (x) = exp(−ikx) + x
These are called the Jost solutions, which have the following property: lim φ±k (x) = exp(±ikx).
x→∞
(9.7.50)
482
9 Solitons and the Inverse Scattering Transform
As already stated earlier, when k is real, the eigenfunctions are continuous (unbounded) states, since λ > 0. On the other hand, if k is purely imaginary, so that k = iκ, where κ is real, then λ < 0, and the eigenfunctions correspond to discrete (bound) states. Henceforth, we adopt k for continuous states and iκ for discrete states. Substituting φ±k (x) in (9.7.47) leads to the inhomogeneous wave equation ∂2K ∂ 2K − = uK, ∂x2 ∂z 2
(9.7.51)
and K(x, z) is related to the potential u(x, t) by u(x, t) = −2
d K(x, x), dx
(9.7.52)
where K(x, z) satisfies the following conditions: lim K(x, z) = 0
z→±∞
and
lim
z→±∞
∂K(x, z) = 0. ∂z
(9.7.53ab)
Because φk (x) and φ−k (x) are linearly independent, they can be treated as fundamental solutions of the original equation (9.7.47). Hence, the general solution of (9.7.47) corresponding to the eigenvalue k can be written as a linear combination of these two solutions. In particular, we examine the solution ψk (x) of (9.7.47) with asymptotic behavior ψk (x) ∼ exp(−ikx), as x → −∞. After some algebraic manipulation, it can be shown that ψk (x) is given by ψk (x) =
1 b(k) φ−k (x) + φk (x). a(k) a(k)
(9.7.54)
Substituting results for φ+k (x) from (9.7.48) and (9.7.49) in (9.7.54) yields
∞
K(x, z) exp(−ikz) dz a(k)ψk (x) = exp(−ikx) + x ∞ + b(k) exp(ikx) + K(x, z) exp(ikz) dz . (9.7.55) x
Multiplying this equation by (2π)−1 exp(iky), where y > x, and integrating with respect to k gives the following result: ∞ 1 a(k)ψk (x) exp(iky) dk 2π −∞ ∞
1 exp ik(y − x) dk = 2π −∞ ∞ ∞
1 K(x, z) exp ik(y − z) dk dz + 2π −∞ x
9.7 The Inverse Scattering Transform (IST) Method
∞
1 b(k) exp ik(x + y) dk 2π −∞ ∞ ∞
1 K(x, z) b(k) exp ik(y + z) dk dz + 2π −∞ x ∞ = K(x, y) + B(x + y) + K(x, z)B(y + z) dz,
483
+
(9.7.56)
x
in which the following results are used: ∞ 1 b(k) exp(ikx) dk B(x) = 2π −∞ and
1 2π
∞ −∞
exp ik(y − z) dk = δ(y − x) = 0,
y > x.
(9.7.57a)
(9.7.57b)
Clearly, the function B(x) represents the Fourier transform of the reflection coefficient b(k) and is a known quantity. The right-hand side of (9.7.55) has terms involving only an unknown K, which is to be determined. In order to determine K, we first simplify the left-hand side of (9.7.56) by using the theory of residues of analytic functions. Without any further details, it turns out from the residue computation from the simple pole of a(k) at k = iκ with ψiκ (x) = cκ φiκ (x) that the left-hand side of (9.7.56) gives ∞ 1 a(k)ψk (x) exp(iky) dk 2π −∞ ∞
= −c2κ exp −κ(x + y) − c2κ K(x, z) exp −κ(y + z) dz. (9.7.58) −∞
There is a corresponding contribution to this equation from every bound state of the potential. Therefore, if there are N discrete states, (9.7.58) has to be modified by including a summation over the individual states. Finally, we combine this with result (9.7.56) and incorporate the t-dependence of each quantity to obtain the integral equation for K(x, y; t) in the form ∞ K(x, z; t)B(y + z; t) dz = 0, (9.7.59) K(x, y; t) + B(x + y; t) + x
where B(x, t) is now given by B(x, t) =
N n=1
c2n (t) exp(−κn x)
1 + 2π
∞
b(k, t) exp(ikx) dk.
(9.7.60)
−∞
Equation (9.7.59) with the kernel B(x, t) given by (9.7.60) is known as the Gelfand–Levitan–Marchenko (GLM) linear integral equation. The solution K(x, y; t) of this integral equation is related to the potential u(x, t) by the following result:
484
9 Solitons and the Inverse Scattering Transform
u(x, t) = −2
d K(x, x; t). dx
(9.7.61)
This is the exact solution of the original KdV equation. However, it does not admit any physical interpretation; that can be achieved from the following asymptotic analysis and examples. For future reference, it is convenient to write the final explicit formula for B(x, t) as B(x, t) =
N
c2n (0) exp 8κ3n t − κn x
n=1
+
1 2π
∞
b(k, 0) exp 8ik 3 t + ikx dk.
(9.7.62)
−∞
In general, it is not possible to solve the integral equation (9.7.59) except for the reflectionless potentials (b(k) = 0). For the general case, the longtime solution represents N solitons traveling with various speeds to the right, and an oscillatory wavetrain with amplitude decreasing with time can be found to follow the series of solitons. For a given initial potential energy function, the eigenvalues κn are constants for all time. Physically, this means that bound-state energy levels are completely specified and remain unchanged while varying the potential through the KdV equation. Another remarkable fact to observe is that the number of solitons that eventually develops is exactly the number of bound states. Of course, this number depends on the initial state u(x, 0) = u0 (x), which is sufficient for determining the number of developing solitons. This completes the description of the inverse scattering transform method for solving the KdV equation. In summary, two distinct steps are involved in the method: (i) the solution of the Schrödinger equation (the Sturm–Liouville problem) for a given initial condition, u(x, 0) = u0 (x), from which we determine the scattering data S(t), and (ii) the solution of the GLM linear integral equation. Even though these two steps may not be technically easy to handle, in principle, the problem is solved. The effectiveness of the method can be best exemplified by many simple but nontrivial examples. The power and success of the inverse scattering transform method for solving the KdV equation can be attributed to several facts. First, the most remarkable result of the method is the fact that the discrete eigenvalues of the Schrödinger equation do not change as the potential evolves according to the KdV equation. Second, the method has reduced solving a nonlinear partial differential equation to solving two linear problems: (i) a second-order ordinary differential equation and (ii) an integral equation. Third, the eigenvalues of the ordinary differential equation are constants, and this leads to a major simplification in the evolution equation for ψ. Fourth, the time evolution of the scattering data is explicitly determined from the asymptotic form of ψ, as |x| → ∞. So, this information allows us to solve the inverse scattering problem, and hence, to find the final solution of the KdV equation. The method is presented schematically in Figure 9.10.
9.7 The Inverse Scattering Transform (IST) Method
485
Fig. 9.10 The IST method for the KdV equation.
We now illustrate the method for the reflectionless case (b(k) = 0) with initial profiles given by the following examples. Example 9.7.1 (Single-Soliton Solution). We consider a particular case where the potential has a single bound state with the eigenvalue λ = −κ21 , normalized constant c0 , and with zero reflection coefficient (b(k, t) ≡ 0) for all continuous states k. In this case, B is obtained from (9.7.62) in the form B(x, t) = c20 (0) exp 8κ31 t − κ1 x = c20 exp 8κ31 t − κ1 x . (9.7.63) The associated GLM equation (9.7.60) can be solved by assuming a separable kernel K(x, y; t) =
N
Kn (x, t) exp(−κn y),
(9.7.64)
n=1
which, in this case (N = 1), becomes K(x, y; t) = K1 (x, t) exp(−κ1 y).
(9.7.65)
The GLM equation (9.7.60) with (9.7.65) is given by
K(x, y; t) + c20 exp 8κ31 t − κ1 (x + y) 3 ∞ 2 + c0 exp 8κ1 t − κ1 y K(x, z; t) exp(−κ1 z) dz = 0. (9.7.66) x
Substituting (9.7.65) in (9.7.66) and performing the integration over z leads to an algebraic equation for K1 (x, t),
K1 (x, t) exp(−κ1 y) + c20 exp 8κ31 t − κ1 (x + y) c2 (9.7.67) + 0 exp 8κ31 t − κ1 y − 2κ1 x K1 (x, t) = 0. 2κ1 This equation solved for K1 (x, t) can be written in the form K1 (x, t) = −κ1 exp 4κ31 t + κ1 x0 sech κ1 x − x0 − 4κ21 t ,
(9.7.68)
486
9 Solitons and the Inverse Scattering Transform
where x0 is defined by x0 = (2κ1 )−1 log
c20 . 2κ1
(9.7.69)
We substitute (9.7.68) in (9.7.65) to obtain the solution for K(x, y; t). Consequently, the solution (9.7.61) of the KdV equation becomes u(x, t) = −2κ21 sech2 κ1 (x − U t) − κ1 x0 , (9.7.70) where U = 4κ21 . This represents a single-soliton solution of amplitude −2κ21 traveling to the right with speed U = 4κ21 and centered initially at the point x0 . Example 9.7.2 (N-Soliton Solution for a sech2 Potential). We investigate the scattering problem for the class of sech2 potentials given by u(x, 0) = u0 (x) = −U sech2 x,
(9.7.71)
where U is a constant. The associated Schrödinger equation is ψxx + λ + U sech2 x ψ = 0.
(9.7.72)
We introduce a change of variable by y = tanh x so that −1 < y < 1 for −∞ < x < ∞. In terms of y, the potential (9.7.71) becomes U sech2 x = U 1 − tanh2 x = U 1 − y 2 , and the second derivative in terms of y is given by d2 dy d dy d 2 d 2 d = 1 − y . 1 − y = dx2 dx dy dx dy dy dy Consequently, equation (9.7.72) gives the associated Legendre equation dψ d λ ψ = 0. 1 − y2 + U+ dy dy 1 − y2
(9.7.73)
We set U = N (N + 1), where N is a positive integer, and then, consider only the bound states (b(k, t) = 0 for all k). Physically, this means that the incident wave is totally reflected, and hence, the associated potential is referred to as reflectionless. If λ = −κ2 < 0, then only bounded solutions for −1 ≤ y ≤ 1 occur when κn = n, n = 1, 2, . . . , N , and the corresponding discrete eigenfunctions are given by ψn (x) = An PNn (tanh x), where the associated Legendre functions PNn (y) are defined by n/2 dn PN (y) PNn (y) = (−1)n 1 − y 2 dy n
and
PN (y) =
N 1 dN 2 y −1 . N N N !2 dy (9.7.74ab)
9.7 The Inverse Scattering Transform (IST) Method
487
PN (y) is the Legendre polynomial of degree N , and the constants of proportionality An are determined from the normalization condition (9.7.24). The asymptotic representation of the discrete eigenfunctions has the form ψn (x) ∼ cn (t) exp(−nx) where
as x → +∞,
cn (t) = cn (0) exp 4n3 t
(9.7.75) (9.7.76)
and cn (0) are also determined from the normalization condition. The function B(x; t) involved in the GLM integral equation (9.7.60), therefore, is given by N c2n (0) exp 8n3 t − nx . (9.7.77) B(x; t) = n=1
Finally, the GLM equation can easily be solved, and hence, the N -soliton solution assumes the asymptotic form u(x, t) ∼ −
N
2n2 sech2 n x − 4n2 t ∓ εn
as t → ±∞,
(9.7.78)
n=1
where the quantities εn are known phases. Clearly, a reflectionless potential with N bound states corresponds to pure N solitons ordered according to their amplitudes. As t → ∞, the tallest (hence, the fastest) soliton is at the front, followed by a series of progressively shorter (therefore, slower) ones behind. All N solitons interact at t = 0 to form the single sech2 profile, which was prescribed as the initial condition at that instant. In particular, when N = 1, U = 2 and κ = κ1 = 1. The corresponding eigenfunction is given by ψ1 (x) = A1 P11 (tanh x) = −A1 sech x, 2 where the constant √ A1 is determined from the condition (9.7.24), which gives 2A = 1 or A = ±(1/ 2). So, the asymptotic nature of the above solution is √ ψ1 (x) ∼ 2e−x as x → ∞. √ √ Therefore, c1 (0) = 2, and c1 (t) = 2 exp(4t), and the function B(x; t) = 2 exp(8t − x). Finally, the associated GLM equation ∞ K(x, z; t) + 2 exp 8t − (x + z) + 2 K(x, y; t) exp 8t − (y + z) dy = 0 x
gives the solution K(x, z; t) = F (x; t) exp(−z), for some function F (x; t) that satisfies the equation
488
9 Solitons and the Inverse Scattering Transform
F + 2 exp(8t − x) + 2F exp(8t)
∞
e−2y dy = 0.
x
This can readily be solved to find the function F expressed as F (x; t) =
−2e8t−x . 1 + e8t−2x
Thus, the final solution is given by 8e2x−8t 2e8t−2x ∂ = − = −2 sech2 (x − 4t). u(x, t) = 2 ∂x 1 + e8t−2x (1 + e2x−8t )
(9.7.79)
This represents the single-soliton solution of amplitude −2 and speed 4. Similarly, the two-soliton solution can be obtained from the initial profile (9.7.71) with N = 2, so that (9.7.80) u(x, 0) = −6 sech2 x. This initial condition is then evolved in time according to the KdV equation, and the solution consists of two solitons where the taller one catches up to the shorter one. They undergo a nonlinear interaction according to the KdV equation and, then, emerge from the interaction unchanged in waveform and amplitude. Eventually, the taller soliton reappears to the right and passes away from the shorter one as t increases. The wave profile u(x, t) with the initial condition (9.7.80) is plotted in Figure 9.11 as a function of x for different values of time: (a) t = −0.5, (b) t = −0.1, (c) t = 0.0, (d) t = 0.1, and (e) t = 0.5. Similarly, the three-soliton solution (N = 3) with u(x, 0) = −12 sech2 x is shown in Figure 9.12 for different values of t: (a) t = 0.0, (b) t = 0.05, and (c) t = 0.2, where evolving solitons have amplitudes 18, 8, and 2, respectively. Finally, to obtain the soliton solution, we examine the asymptotic behavior of the exact solution u(x, t) for large x and t by considering only B(x, t) given by (9.7.62). For large t and x ∼ 4κ2N t, where κN is the largest eigenvalue (κ1 < κ2 < · · · < κN ), the N th term in the series in (9.7.62) dominates over all other terms, and the significant contribution to the integral in (9.7.62) dominates over all other terms. This contribution to the integral in (9.7.62) can be found from the stationary phase approximation (Segur 1973) in the form t−1/3 as t → ∞. Thus, it turns out that B(x, t) ∼ α exp(−κN x), where
(9.7.81)
α = c2N (0) exp 8κ3N t .
With this value of B and y = x, the associated GLM equation becomes K(x, x, t) + α exp(−2κN x) ∞ + α exp(−κN x) exp(−κN z)K(x, z, t) dz ∼ 0. x
This can be easily solved by writing
(9.7.82)
9.7 The Inverse Scattering Transform (IST) Method
Fig. 9.11 The two-soliton solution. From Drazin and Johnson (1989).
Fig. 9.12 The three-soliton solution. From Drazin and Johnson (1989).
489
490
9 Solitons and the Inverse Scattering Transform
K(x, z, t) = F (x, t) exp(−κN z), where F satisfies
−κN x
F +α e
∞
+F
−2κN z
e
(9.7.83)
dz = 0.
(9.7.84)
x
Result (9.7.84) gives
F = −α exp(−κN x)
α 1+ exp(−2κN x) . 2κN
(9.7.85)
Substituting (9.7.83) in (9.7.61) gives u(x, t) ∼ −2κ2N sech2 κN (x − x0 ) − 4κ3N t ,
(9.7.86)
where c2N (0)/2κN = exp(2κN x0 ). Obviously, (9.7.86) asymptotically represents a soliton of amplitude 2κ2N and velocity 4κ2N . If, instead of large t, we consider x ∼ 4κ2N t for any arbitrary n, the significant term in the series involved in (9.7.86) is the nth term, and then, a soliton of amplitude 2κ2n and speed 4κ2n would emerge. Thus, the upshot of this analysis is that the initial (potential well) profile disintegrates into N solitons corresponding to discrete eigenvalues of the associated Schrödinger equation. A further elaborate mathematical analysis of the integral in (9.7.62) reveals a complete solution consisting of a series of N solitons preceded by an oscillatory trail of amplitude decreasing with time. This disintegration process of an initial profile into a series of solitons is usually called fission. Examples 9.7.1 and 9.7.2 illustrate the reflectionless (b(k) = 0) initial profiles. We now give examples for the nonzero reflection coefficient (b(k) = 0). Example 9.7.3 (A Soliton Solution Associated with the Dirac Delta Function Initial Profile). In this case, we use the initial condition as u(x, 0) = −u0 δ(x),
(9.7.87)
where u0 is a positive constant. For a discrete eigenstate, λ1 = −κ21 , where κ1 = 1 2 u0 and the corresponding eigenfunction is given by √ κ1 exp(−κ1 x) if x > 0, (9.7.88ab) ψ1 (x) = √ κ1 exp(+κ1 x) if x < 0. √ The continuous eigenfunction for k = λ exists and can be written as b(k) = −
u0 . (u0 + 2ik)
(9.7.89)
We can then use the time evolution of the scattering data c1 (t) =
√ κ1 exp 4κ31 t
and
b(k, t) = −
u0 exp 8ik 3 t . (9.7.90ab) (u0 + 2ik)
9.7 The Inverse Scattering Transform (IST) Method
491
Fig. 9.13 (a) Initial profile and (b) solution at a later time.
The function B(x, t) can be represented as u0 B(x, t) = κ1 exp 8κ31 t − κ1 x − 2π
∞
−∞
exp(8ik3 t + ikx) dk. (u0 + 2ik)
(9.7.91)
However, the function K(x, y; t) cannot easily be determined from (9.7.59). But the asymptotic solution for u(x, t) for the single soliton associated with the discrete eigenvalue κ1 = 12 u0 and fixed (x − u20 t) is given by 1 1 u(x, t) ∼ − u20 sech2 u0 x − u20 t − ε1 , as t → ∞, (9.7.92) 2 2 where the phase change is given by exp(2κ1 ε1 ) = c20 (0)/2κ1 . In fact, the asymptotic solution (9.7.92) is made up of the contribution to B(x, t) from the first term in (9.7.91), and the integral term vanishes where asymptotic solution (9.7.92) is valid. The initial profile represented by the delta function is shown in Figure 9.13(a), and the solution u(x, t) for a subsequent time is plotted in Figure 9.13(b). Finally, if the amplitude u0 of the initial profile is negative, there is no discrete eigenvalue, and hence, there is no soliton. And only dispersive waves exist in the solution for t > 0. As shown above, if u0 > 0, there is only one eigenvalue κ1 = u20 , and hence, a single soliton with amplitude 2κ21 = 12 u20 is generated. Example 9.7.4 (Solitons Associated with Negative sech2 Initial Profiles). We use the method described by Crandall (1991) to examine the development of solitons and a dispersive wave associated with a class of potentials given by u(x, 0) = u0 (x) = −U sech2 x.
(9.7.93)
492
9 Solitons and the Inverse Scattering Transform
Fig. 9.14 The time evolution of the solution at three different times. From Drazin and Johnson (1989).
According to the general theory of solitons, the total number of discrete eigenvalues for general U is given by 1 1 1 2 + 1, − ℵ= U+ 4 2 1
(9.7.94)
where [·] denotes the integral part and, if {(U + 14 ) 2 − 12 } is an integer, then the +1 is omitted. For U = 4, ℵ = 2, and hence, there are two discrete eigenvalues, and therefore, a solution with two solitons. Moreover, U = 4 cannot be written as N (N + 1) for integral N , and the solution also includes a dispersive wave. Finally, the solution of the KdV equation is determined by numerical integration. The end result is that there are contributions from both the discrete and continuous states to the function B(x + t) in equation (9.7.60). The effect of the different contributions is seen in the time evolution of the solution, which reveals the appearance of two solitons moving to the right and a dispersive wavetrain traveling to the left, as shown in Figure 9.14. We close this section by adding a brief discussion on the number of solitons associated with an initial profile u(x, 0) = u0 (x). The solution depends on the discrete eigenvalues of the associated Schrödinger equation. We discuss this point by citing some specific examples.
9.7 The Inverse Scattering Transform (IST) Method
493
For the case u0 (x) = −a sech2 (bx), the number of eigenvalues is given by (Landau and Lifshitz 1959, p. 70) 1 κn = b 2
1/2 4a 1+ 2 − (2n − 1) ≥ 0, b
(9.7.95)
where the number of eigenvalues is determined by the parameter P defined by ∞ u0 (x)1/2 dx. (9.7.96) P = −∞
√ For the sech2 profile, P = πb a. The number of solitons is given by N = largest integer ≤
1 2
1+
4P 2 π2
1/2
+1 .
(9.7.97)
This shows that the number √ of solitons depends on the parameter P , which, in this case, is proportional to b a. For the delta function case, n exp −nx2 , (9.7.98) u0 (x) = −aδ(x) = −a lim n→∞ π and P → 0 as n → ∞, and hence, there is only one soliton consistent with (9.7.97). In the other limit as P → ∞, the formula (9.7.97) gives N∼
P . π
(9.7.99)
Thus, the result (9.7.97) shows that there is always one soliton for small P , and as P increases (P → ∞), the number of solitons increases. Furthermore, when the initial disturbance is large (P → ∞), there are many closely spaced eigenvalues that satisfy the famous Bohr–Sommerfeld rule 1/2 1 . (9.7.100) p dx = λ − u0 (x) dx = 2π n + 2 Thus, the number of solitons (the largest value of n for λ = 0) is given by 1/2 P 1 ∞ N∼ (9.7.101) u0 (x) dx = . π −∞ π This ensures the validity of the result obtained in the previous examples. Another approximate formula for the number of solitons with amplitudes in (a, a + da) was first obtained by Karpman (1967) in the form 1 N (a) = 8π
u0 (x) − a 2
−1/2 dx.
(9.7.102)
494
9 Solitons and the Inverse Scattering Transform
This is over the range 0 < a < 2u0m = 2|u0 |max , where u0m is the largest value of √ |λ| for the bound states in (9.7.100), since the range of κ is 0 < κ < u0m . Hence, the total number of solitons is given by 2u0m 1 N (a) da = (9.7.103) N= |u0 |1/2 dx. π 0 This is in agreement with (9.7.101). Thus, the dependence of P on the size (a and b) of a soliton suggests another general result: ∞ |u|1/2 dx. (9.7.104) Q= −∞
Physically, this represents an interesting measure of soliton √ shapes for a single soliton, and (9.7.104) can easily be computed to obtain Q ∼ 2π, √ which is independent of the amplitude κN . Since solitons here are of unit size, Q = 2πN for a series of N solitons. This shows that there is a Planck constant for solitons! It is worth noting here that the parameter P is the value of the integral in the initial disturbance. For large P , N ∼ P/π, and for large time t, √ √ (9.7.105) Q = 2πN ∼ 2P. This describes the close connection between the initial P and the final Q.
9.8 Bäcklund Transformations and the Nonlinear Superposition Principle Historically, Bäcklund transformations were developed in the 1880s to study the related theories of differential geometry and differential equations. They occurred as an extension of contact transformations, which transform surfaces with a common tangent at a point in one space into surfaces in another space, which also have a common tangent at the corresponding point. One of the earliest Bäcklund transformations was found for the sine-Gordon equation, uxt = sin u. This equation originally arose in differential geometry in connection with the theory of surfaces of constant negative curvature. However, the study of these transformations had been dormant until the recent work on solitons in the 1970s. It has been recognized that partial differential equations admit soliton-like solutions if and only if they admit Bäcklund transformations. Indeed, there is a close relationship between the inverse scattering transform (IST) and a Bäcklund transformation (BT), in the sense that the scattering problem and the associated time dependence that constitute an IST also constitute a BT. In other words, every evolution equation solvable by an IST has a corresponding BT, and, conversely, the existence of BTs always or almost always implies integrability by the IST. Another approach to deriving conservation laws and the inverse scattering problem is through the use of the Bäcklund transformations. For second-order partial
9.8 Bäcklund Transformations and the Nonlinear Superposition Principle
495
differential equations, the Bäcklund transformation consists of a pair of first-order partial differential equations relating the solutions of the given equation to another solution of the same equation or to a solution of another second-order equation. In general, a Bäcklund transformation for a second-order partial differential equation for a dependent variable u(x, t) is defined by the pair of equations
wx = P (w, u, ux , ut , x, t), (9.8.1ab) wt = Q(w, u, ux , ut , x, t), where P and Q are functions of the variables indicated, but not of the derivatives of w. A new equation for w is obtained from the consistency condition. One of the simplest Bäcklund transformations is the pair of the Cauchy–Riemann equations in complex analysis ux = vy
and
uy = −vx ,
(9.8.2ab)
for the Laplace equations uxx + uyy = 0 and
vxx + vyy = 0.
(9.8.3ab)
Thus, if v(x, y) = xy is a simple solution of the Laplace equation, then u(x, y) can be determined from ux = x and uy = −y. Therefore, u(x, y) = 12 (x2 − y 2 ) is another solution of the Laplace equation. In connection with the Miura transformation (9.6.10), it has been shown that, if v is a solution of the modified KdV equation (9.6.11), then u is a solution of the KdV equation (9.6.3). Since it is possible to eliminate higher derivatives from (9.6.11) by using (9.6.10), we can treat (9.6.10) and (9.6.11), written in the form vx = u − v 2
and vt = 6v 2 vx − vxxx ,
(9.8.4ab)
as a Bäcklund transformation for the KdV equation (9.6.3). Wahlquist and Estabrook (1973, 1975, 1976) have developed a more convenient and useful Bäcklund transformation for solutions of the KdV equation. Since the KdV equation is Galilean invariant, we replace v in (9.6.11) by u − λ, where λ is a real parameter. Using u = λ + v 2 + vx , we rewrite the modified KdV equation (9.6.11) in the form (9.8.5) vt − 6 v 2 + λ vx + vxxx = 0, so that u satisfies the KdV equation (9.6.3). Clearly, both v and −v satisfy (9.8.5). This leads us to construct two functions u1 and u2 , corresponding to v and −v, in the form u 2 = λ + v 2 − vx (9.8.6ab) u 1 = λ + v 2 + vx , for a given λ and v. Consequently, u1 − u2 = 2vx ,
u1 + u2 = 2 λ + v 2 .
(9.8.7ab)
496
9 Solitons and the Inverse Scattering Transform
It is convenient to define a new potential function wi such that ui = ∂wi /∂x, i = 1, 2. Thus, equations (9.8.7ab) reduce to the following pair of equations: w1 − w2 = 2v,
(9.8.8)
1 (w1 + w2 )x = 2λ + (w1 − w2 )2 , 2
(9.8.9)
where any arbitrary function of t in (9.8.8) has been incorporated in wi without changing ui . We next use (9.8.7ab) and (9.8.8) in order to transform (9.8.5) into the form 2 2 − (w1 − w2 )xxx . − w2x (9.8.10) (w1 − w2 )t = 3 w1x Thus, the upshot of this analysis is that equations (9.8.9) and (9.8.10) constitute a pair of auto-Bäcklund transformations of the KdV equation, where the former is the x part and the latter is the t part. We next illustrate the method by solving (9.8.9), (9.8.10) for w1 with w2 = 0 for all x and t. The resulting equations for w1 become 1 w1x = 2λ + w12 and 2 2 w1t = 3w1x − w1xxx ,
(9.8.11a) (9.8.11b)
where the former gives w1xxx =
∂ 2 + w12 w1x , (w1 w1x ) = w1x ∂x
(9.8.12)
and hence, (9.8.11b) can be written, by using (9.8.11a) again and λ = κ2 , as w1t + 4κ2 w1x = 0.
(9.8.13)
This equation admits the general solution w1 (x, t) = f x − 4κ2 t ,
(9.8.14)
where f is an arbitrary function. On the other hand, (9.8.11a) can readily be integrated to obtain the solution (9.8.15) w1 (x, t) = −2κ tanh κx + α(t) , where α(t) is an arbitrary function of t. For consistency of the solutions (9.8.14) and (9.8.15), we require that α(t) = −4κ(κ2 t − x0 ), where x0 is an arbitrary constant. Thus, the Bäcklund transformations give the final solution (9.8.16) w1 (x, t) = −2κ tanh κ x − x0 − 4κ2 t , and hence, it follows from u1 = w1x that
u1 (x, t) = −2κ2 sech2 κ x − x0 − 4κ2 t .
This is the soliton solution of the KdV equation (9.6.3).
(9.8.17)
9.8 Bäcklund Transformations and the Nonlinear Superposition Principle
497
All of this illustrates how the Bäcklund transformations can be used to obtain the soliton solution of the KdV equation. However, the procedure just demonstrated requires two integrations, one with respect to x and the other with respect to t. Instead of the process of integration involving arbitrary functions, Wahlquist and Estabrook (1973) developed an elegant method, based on the theory of differential forms, for determining solutions of the KdV equation to obtain a one-parameter family of Bäcklund transformations. In this method, two distinct solutions w1 and w2 are generated from the Bäcklund transformations by using the same given solution w with two different values λ1 and λ2 of λ. Thus, equation (9.8.9) can be written in two different forms, 1 (9.8.18a) (w1 + w)x = 2λ1 + (w1 − w)2 , 2 1 (w2 + w)x = 2λ2 + (w2 − w)2 . (9.8.18b) 2 It is now possible to construct another solution w12 from w1 and λ2 and, similarly, a solution w21 from w2 and λ1 , so that 1 (9.8.19a) (w12 + w1 )x = 2λ2 + (w12 − w1 )2 , 2 1 (w21 + w2 )x = 2λ1 + (w21 − w2 )2 . (9.8.19b) 2 We next use Bianchi’s theorem of permutability for the Bäcklund transformations in differential geometry, which states that w12 = w21 . Now, we subtract the difference of equations (9.8.18ab) from the difference of equations (9.8.19ab) and use the identity w12 = w21 , so as to produce zero on the left-hand side of the resulting equation: 1 (w12 − w1 )2 − (w21 − w2 )2 − (w1 − w)2 + (w2 − w)2 , 2 whence the solution for w12 (= w21 ) is given by 0 = 4(λ2 − λ1 ) +
w12 = w −
4(λ1 − λ2 ) . (w1 − w2 )
(9.8.20)
Thus, it is now possible to find solutions of the KdV equation in a straightforward manner. Equation (9.8.20) is a purely simple algebraic expression, known as the nonlinear superposition principle, for constructing solutions. It is possible to generalize superposition formula (9.8.20) and use the procedure to construct multisoliton solutions. A process similar to the two-soliton solution can be generalized to obtain a threesoliton solution of the KdV equation: 4(λ2 − λ3 ) w12 − w13 λ1 w1 (w2 − w3 ) + λ2 w2 (w3 − w1 ) + λ3 w3 (w1 − w2 ) . = λ1 (w2 − w3 ) + λ2 (w3 − w1 ) + λ3 (w1 − w2 )
w123 = w1 −
(9.8.21)
This represents the nonlinear superposition principle for the three-soliton solution.
498
9 Solitons and the Inverse Scattering Transform
Thus, the multisoliton can be produced by continuing this process and setting √ (9.8.22) ws = −4κ2s sech2 2κs x − 8κ2s t − xs , where xs is constant. In particular, when w = 0, w1 and w2 represent bounded and unbounded solutions corresponding to bounded and unbounded solitons for u expressed as √ u1 = −4κ2 sech2 2κ x − 8κ2 t − x0 (9.8.23) and u2 = −4κ2 cosech2
√ 2κ x − 8κ2 t − x0 .
(9.8.24)
Although solution (9.8.24) is not of physical interest, it is essential for the construction of a bounded solution for w12 , which, in the present case, represents a two-soliton solution. Example 9.8.1 (Two-Soliton Solution). We apply the nonlinear superposition principle (9.8.20) to derive the two-soliton solution of the KdV equation. We take w0 = 0,
w1 = −2 tanh(x − 4t),
w2 = −4 coth(2x − 32t),
(9.8.25)
so that λ1 = 1 and λ2 = −4. Consequently, (9.8.20) becomes w12 = −
6 . {2 coth(2x − 32t) − tanh(x − 4t)}
(9.8.26)
The corresponding solution of the KdV equation follows from the result u12 =
∂ 6{4cosech2 (2x − 32t) + sech2 (x − 4t)} w12 = − ∂x {2 coth(2x − 32t) − tanh(x − 4t)}2
6{4 cosh2 (x − 4t) + sinh2 (2x − 32t)} {2 cosh(2x − 32t) cosh(x − 4t) − sinh(2x − 32t) sinh(x − 4t)}2 12{3 + 4 cosh(2x − 8t) + cosh(4x − 64t)} =− . (9.8.27) {3 cosh(x − 28t) + cosh(3x − 36t)}2 =−
This represents the two-soliton solution associated with the initial profile (9.7.80). The numerical values of (9.8.27) for five different times: (a) t = −0.5, (b) t = −0.1, (c) t = 0.0, (d) t = 0.1, and (e) t = 0.5 plotted as a function of x give the same wave profile u(x, t) as in Figure 9.11. We close this section by citing some related work on the use of Bäcklund transformations in finding N -soliton solutions of several nonlinear evolution equations in Wronskian form. Several authors, including Freeman (1984) and Nimmo and Freeman (1983) used this method to obtain N -soliton solutions of KdV, modified KdV, sine-Gordon, KP, Boussinesq, nonlinear Schrödinger equations, and other partial differential difference equations. However, it seems that the scope of this approach is still not yet fully understood.
9.9 The Lax Formulation and the Zakharov and Shabat Scheme
499
9.9 The Lax Formulation and the Zakharov and Shabat Scheme In his 1968 seminal paper, Lax developed an elegant formalism for finding isospectral potentials as solutions of a nonlinear evolution equation with all of its integrals. This work deals with some new and fundamental ideas and deeper results and their application to the KdV model. This work subsequently paved the way to generalizations of the technique as a method for solving other nonlinear partial differential equations. Introducing the Heisenberg picture, Lax developed the method of inverse scattering based upon an abstract formulation of evolution equations and certain properties of operators on a Hilbert space, some of which are familiar in the context of quantum mechanics. His formulation has the feature of associating certain nonlinear evolution equations with linear equations which are analogs of the Schrödinger equation for the KdV equation. To formulate Lax’s method, we consider two linear operators L and M . The eigenvalue equation related to the operator L corresponds to the Schrödinger equation for the KdV equation. The general form of this eigenvalue equation is Lψ = λψ,
(9.9.1)
where ψ is the eigenfunction and λ is the corresponding eigenvalue. The operator M describes the change of the eigenvalues with the parameter t, which usually represents time in a nonlinear evolution equation. The general form of this evolution equation is (9.9.2) ψt = M ψ. Differentiating (9.9.1) with respect to t gives Lt ψ + Lψt = λt ψ + λψt .
(9.9.3)
We next eliminate ψt from (9.9.3) by using (9.9.2) and obtain Lt ψ + LM ψ = λt ψ + λM ψ = λt ψ + M λψ = λt ψ + M Lψ,
(9.9.4)
or equivalently, ∂L ψ = λt ψ + (M L − LM )ψ. (9.9.5) ∂t Thus, eigenvalues are constant for nonzero eigenfunctions if and only if ∂L = −(LM − M L) = −[L, M ], ∂t
(9.9.6)
where [L, M ] = (LM − M L) is called the commutator of the operators L and M , and the derivative on the left-hand side of (9.9.6) is to be interpreted as the time derivative of the operator alone. Equation (9.9.6) is called the Lax equation and the operators L and M are called the Lax pair. It is the Heisenberg picture of the KdV equation. The problem, of course, is how to determine these operators for a given evolution equation. There is no systematic method of solution of this problem. For
500
9 Solitons and the Inverse Scattering Transform
a negative integrable hierarchy and in order to find a Lax pair from a given spectral problem, Qiao (1995) and Qiao and Strampp (2002) suggested a general approach to generate integrable equations. We consider the initial-value problem for u(x, t) which satisfies the nonlinear evolution equation system ut = N (u), u(x, 0) = f (x),
(9.9.7) (9.9.8)
where u ∈ Y for all t, Y is a suitable function of space, and N : Y → Y is a nonlinear operator that is independent of t but may involve x or derivatives with respect to x. We must assume that the evolution equation (9.9.7) can be expressed in the Lax form (9.9.9) Lt + (LM − M L) = Lt + [L, M ] = 0, where L and M are linear operators in x on a Hilbert space H and depend on u and Lt = ut is a scalar operator. We also assume that L is self-adjoint so that (Lφ, ψ) = (φ, Lψ) for all φ and ψ ∈ H with (·, ·) as an inner product. We now formulate the eigenvalue problem for ψ ∈ H: Lψ = λ(t)ψ,
t ≥ 0, x ∈ R.
(9.9.10)
Differentiating with respect to t and making use of (9.9.9), we obtain λt ψ = (L − λ)(ψt − M ψ). The inner product of ψ with this equation yields (ψ, ψ)λt = (L − λ)ψ, λt − M ψ ,
(9.9.11)
(9.9.12)
which is, since L − λ is self-adjoint, given by (ψ, ψ)λt = (0, ψt − M ψ) = 0. Hence, λt = 0, confirming that each eigenvalue of L is a constant. Consequently, (9.9.11) becomes (9.9.13) L(ψt − M ψ) = λ(ψt − M ψ). This shows that (ψt − M ψ) is an eigenfunction of the operator L with the eigenvalue λ. It is always possible to redefine M by adding the product of the identity operator and a suitable function of t, so that the original equation (9.9.9) remains unchanged. This leads to the time evolution equation for ψ as ψt = M ψ, Thus, we have the following.
t ≥ 0.
(9.9.14)
9.9 The Lax Formulation and the Zakharov and Shabat Scheme
501
Theorem 9.9.1. If the evolution equation (9.9.7) can be expressed as the Lax equation (9.9.15) Lt + [L, M ] = 0, and if (9.9.10) holds, then λt = 0 and ψ satisfies (9.9.14). It is not yet clear how to find the operators L and M that satisfy the preceding conditions. To illustrate the Lax method, we choose the Schrödinger operator L in the form ∂2 (9.9.16) L ≡ − 2 + u, ∂x so that Lψ = λψ becomes the Sturm–Liouville problem for the self-adjoint operator L. With this given L, the problem is to find the operator M . Based on the theory of a linear unitary operator on a Hilbert space H, the linear operator M can be chosen as antisymmetric, so that (M φ, ψ) = −(φ, M ψ) for all ψ, φ ∈ H. So, a suitable linear combination of odd derivatives in x is a natural choice for M . It follows from the inner product that ∞ ∞ n ∂nψ ∂ φ ψ dx = − φ n dx = −(φ, M ψ), (9.9.17) (M φ, ψ) = n −∞ ∂x −∞ ∂x provided M = ∂ n φ/∂xn for odd n, and φ, ψ, and their derivatives with respect to x tend to zero, as |x| → ∞. Moreover, we require that M has sufficient freedom in any unknown constants or functions to make Lt + [L, M ] a multiplicative operator, that is, of degree zero. For n = 1, the simplest choice for M is M = c(∂/∂x), where c is a constant. It then follows that [L, M ] = −cux , which is automatically a multiplicative operator. Thus, the Lax equation is Lt + [L, M ] = ut − cux = 0,
(9.9.18)
and hence, the one-dimensional wave equation ut − cux = 0
(9.9.19)
has an associated eigenvalue problem with the eigenvalues that are constants of motion. The next natural choice is M = −a
∂3 ∂ ∂ +A + A + B, ∂x3 ∂x ∂x
(9.9.20)
where a is a constant, A = A(x, t), and B = B(x, t), and the third term on the righthand side of (9.9.20) can be dropped, but we retain it for convenience. It follows from an algebraic calculation that [L, M ] = auxxx − Axxx − Bxx − 2ux A ∂ ∂2 + (3aux − 4Ax ) 2 . +(3auxx − 4Axx − 2Bx ) ∂x ∂x
502
9 Solitons and the Inverse Scattering Transform
This would be a multiplicative operator if A = 34 au and B = B(t). Consequently, the Lax equation (9.9.15) becomes 3 a ut − auux + uxxx = 0. 2 4
(9.9.21)
This is the standard KdV equation if a = 4. The operator M defined by (9.9.20) reduces to the form ∂ ∂ ∂3 + u + B(t). (9.9.22) M = −4 3 + 3 u ∂x ∂x ∂x Hence, the time evolution equation for ψ can be simplified by using the Sturm– Liouville equation, ψxx − (u − λ)ψ = 0, as ψt = 4(λψ − uψ)x + 3ψx + 3(uψ)x + Bψ = 2(u + 2λ)ψx − ux ψ + Bψ.
(9.9.23)
We close this section by adding several comments. First, any evolution equations solvable by the IST, like the KdV equation, can be expressed in Lax form. However, the main difficulty is that there is no completely systematic method of finding whether or not a given partial differential equation produces a Lax equation and, if so, how to find the Lax pair L and M . Indeed, Lax proved that there is an infinite number of operators, M , one associated with each odd order of ∂/∂x, and hence, an infinite family of flows ut under which the spectrum of L is preserved. Second, it is possible to study other spectral equations by choosing alternative forms for L. Third, the restriction that L and M should be limited to the class of scalar operators could be removed. In fact, L and M could be matrix operators. The Lax formulation has already been extended for such operators. Fourth, Zakharov and Shabat (1972, 1974) published a series of notable papers in this field extending the nonlinear Schrödinger (NLS) equation and other evolution equations. For the first time, they have also generalized the Lax formalism for equations with more than one spatial variable. This extension is usually known as the Zakharov and Shabat (ZS) scheme which, essentially, follows the Lax method and recasts it in a matrix form, leading to a matrix Marchenko equation. Finally, we briefly discuss the ZS scheme for non-self-adjoint operators to obtain N -soliton solutions for the NLS equation. Zakharov and Shabat introduced an ingenious method for any nonlinear evolution equation ut = N u,
(9.9.24)
∂L = i[L, M ] = i(LM − M L), ∂t
(9.9.25)
to represent the equation in the form
where L and M are linear differential operators including the function u in the coefficients and L refers to differentiating u with respect to t in the expression for L.
9.9 The Lax Formulation and the Zakharov and Shabat Scheme
503
We consider the eigenvalue problem (9.9.26)
Lφ = λφ. Differentiation of (9.9.26) with respect to t gives ∂λ = (L − λ)(iφt − M φ). iφ dt
(9.9.27)
If φ satisfies (9.9.26) initially and changes in such a manner that (9.9.28)
iφt = M φ,
then φ always satisfies (9.9.26). Equations (9.9.26) and (9.9.28) are the pair of equations coupling the function u(x, t) in the coefficients with a scattering problem. Indeed, the nature of φ determines the scattering potential in (9.9.26), and the evolution of φ in time is given by (9.9.28). Although this formulation is quite general, the crucial step, still, is to factor L according to (9.9.25). Zakharov and Shabat (1972) introduced 2×2 matrices associated with (9.9.25) as follows: 0 u∗ 1+α 0 ∂ , (9.9.29) + L=i u 0 0 1 − α ∂x |u|2 ∗ iu 1 0 ∂2 x + 1+α −|u|2 , M = −α (9.9.30) 0 1 ∂x2 −iux 1−α
and the NLS equation for complex u(x, t) is given by iut + uxx + γ|u|2 u = 0, where
(9.9.31)
γ = 2/ 1 − α2 .
Thus, the eigenvalue problem (9.9.26) and the evolution equation (9.9.28) complete the inverse scattering problem. The initial-value problem for u(x, t) can be solved for a given initial condition u(x, 0). It seems clear that the significant contribution would come from the point spectrum for large times (t → ∞). Physically, the disturbance tends to disintegrate into a series of solitary waves. The mathematical analysis is limited to the asymptotic solutions so that |u| → 0 as |x| → ∞, but a series of solitary waves is expected to be the end result of the instability of wavetrains to modulations. In general, to date, there is no completely systematic method of determining the linear scattering problem associated with a given nonlinear evolution equation. However, one systematic method is the prolongation structure method, which was introduced by Wahlquist and Estabrook (1975, 1976) using the knowledge of Lie algebra. This method was found to be very useful for several nonlinear evolution equations. Dodd and Gibbon (1997) successfully developed the prolongation structure of a higher-order KdV equation. Subsequently, several authors including Dodd and
504
9 Solitons and the Inverse Scattering Transform
Fordy (1983, 1984), Fordy (1990), and Kaup (1980) used the prolongation method for finding the scattering problem of a given nonlinear evolution equation. Finally, we close this section by including another commutative representation of (9.9.9), so that the theory can be extended to other nonlinear evolution equations. It has already been demonstrated that the Schrödinger equation Lψ = k2 ψ
(9.9.32)
plays a fundamental role in solving the KdV equation. It has also been seen that if the potential u in this equation varies in time according to the KdV equation, then ψ satisfies a linear equation: ψt = −4ψxxx + 6uψx + 3ux ψ.
(9.9.33)
A necessary condition for these equations to be compatible for all λ is that u(x, t) satisfies the KdV equation (9.9.1). This can be seen as follows. We first transform the Schrödinger equation (9.9.32) into two first-order equations by introducing a new function φ so that ψx = ikψ + φ
and φx = −ikφ + uψ.
(9.9.34ab)
We can use (9.9.34ab) to eliminate all x-derivatives from (9.9.33), which then reduces to the form ψt = 4ik 3 ψ + 4k2 φ + 2ikψ − ux ψ + 2uφ. Using (9.9.34ab), equation (9.9.35) leads to an equation for φ in the form φt = −4ik 3 φ + 4k 2 uφ + 2ikux ψ − 2ikuφ + 2u2 − uxx ψ + ux φ. Introducing a column vector Ψ =
(9.9.35)
(9.9.36)
ψ , we can reformulate the system of equations φ
(9.9.34ab)–(9.9.36) in a compact form, Ψx = A(x, t; λ)Ψ and Ψt = B(x, t; λ)Ψ,
(9.9.37a) (9.9.37b)
where λ is a complex parameter and A(x, t; λ) and B(x, t; λ) are 2 × 2 matrices in the form iλ 0 0 1 A= + , (9.9.38) 0 −iλ u 0 1 0 0 1 u 0 B = 4iλ3 + 4λ2 + 2iλ 0 −1 u 0 ux −u 2u −ux + , (9.9.39) 2u2 − uxx ux and A and B depend on x and t through u(x, t).
9.9 The Lax Formulation and the Zakharov and Shabat Scheme
505
Differentiating (9.9.37a) with respect to t and (9.9.37b) with respect to x leads to the compatibility equation ∂A(λ) ∂B(λ) − + A(λ), B(λ) = 0. ∂t ∂x
(9.9.40)
This is a new commutative representation which can be used to extend the KdV theory to other nonlinear equations. If we consider the case where A and B are polynomial in λ, then the left-hand side of (9.9.40) is also a polynomial in λ. Since equation (9.9.40) is true for all values of λ, all coefficients of this polynomial equation must be identically zero. We next substitute (9.9.38) and (9.9.39) in (9.9.40) to obtain all coefficients which vanish except for the constant term. This means that the left-hand side of (9.9.40) reduces to the matrix 0 0 , ut − 6uux − uxxx 0 which vanishes, so that u(x, t) satisfies the KdV equation. We illustrate the above method by examples. Example 9.9.1 (Nonlinear Schrödinger Equation). We consider the form of the 2 × 2 matrix as iλ 0 0 iq A= + , (9.9.41) 0 −iλ ir 0 where q = q(x, t) and r = r(x, t) are complex-valued functions of x and t. We chose the matrix B(λ) so that (9.9.40) yields certain partial differential equations for q and r. We assume the matrix B(λ) as rq 0 1 0 0 q 0 qx 2 B = 2iλ −i . (9.9.42) + 2iλ + −rx 0 0 −rq 0 −1 r 0 Substituting these matrices A and B in (9.9.40) gives a system of equations irt + rxx + 2qr2 = 0, iqt + qxx + 2rq 2 = 0.
(9.9.43a) (9.9.43b)
We next write r = q or r = −q in the above equations to obtain irt + rxx ± 2|r|2 r = 0.
(9.9.44ab)
These are known as the nonlinear Schrödinger equations. Example 9.9.2 (The KdV and Modified KdV Equations). We use the same A as in (9.9.41) and choose B as 1 0 0 q rq −iqx 3 2 − 4iλ + 2λ B = −4iλ 0 −1 r 0 irx −rq 2 iqxx + 2irq qrx − rqx , (9.9.45) + irxx + 2iqr 2 −qrx − rqx
506
9 Solitons and the Inverse Scattering Transform
so that equation (9.9.40) becomes (9.9.46a) (9.9.46b)
qt + 6rqqx + qxxx = 0, rt + 6qrrx + rxxx = 0.
We assume that q and r are real functions of x and t. When r = −1, (9.9.46a) is the famous KdV equation. When r = q or r = −q, equations (9.9.46a) and (9.9.46b) give rt ± 6r2 rx + rxxx = 0.
(9.9.47ab)
These are the modified KdV equations. Example 9.9.3 (The sine-Gordon Equation). We choose r = q = matrix A as in (9.9.41), and the matrix B as cos u −i sin u −1 , B(λ) = (4iλ) i sin u − cos u
1 2 ux ,
the same
so that equation (9.9.40) leads to the sine-Gordon equation uxt = sin u.
(9.9.48)
Example 9.9.4 (The sinh-Gordon Equation). We substitute r = −q = − 12 ux , the same matrix A as in (9.9.41), and cosh u −i sinh u , B(λ) = (4iλ)−1 −i sinh u − cosh u in (9.9.40) to obtain the sinh-Gordon equation uxt = sinh u.
(9.9.49)
9.10 The AKNS Method In 1974, Ablowitz, Kaup, Newell, and Segur (AKNS) generalized the ZS scheme so that their method can be applied to solve many other evolution equations. We briefly outline the AKNS scheme below without all the technical details. We begin this discussion by considering the pair of linear equations ux = Au ut = Bu,
and
(9.10.1a) (9.10.1b)
where u is an n-dimensional vector and A and B are n × n matrices. Then differentiating (9.10.1a) with respect to t and (9.10.1b) with respect to x and equating the results leads to the equation
9.10 The AKNS Method
∂A ∂B − + [A, B] = 0. ∂t ∂x
507
(9.10.2)
This is essentially equivalent to the Lax equation (9.9.15). It turns out that, given A, there is a simple deductive method to find a B so that (9.10.2) contains a nonlinear evolution equation. For (9.10.2) to be effective, the operator A should have a parameter which plays the role of an eigenvalue, say ζ, which satisfies the condition ζt = 0. Further, a solution of the related nonlinear evolution equation in an infinite interval can be obtained when the associated scattering problem is such that the inverse scattering method can be carried out. Although there are many nonlinear evolutions which satisfy (9.10.2), at this time, a complete scattering method for many of the associated equations (9.10.1a) has not yet been successfully developed. As an example, we consider a 2 × 2 eigenvalue problem for the pair of equations ψ1x = −iζψ1 + qψ2 , ψ2x = iζψ2 + rψ1 ,
(9.10.3a) (9.10.3b)
that is, where ψ =
ψ1 ψ2
ψx = Aψ, Bψ = −iζψ, (9.10.4ab) −iζ q ∂ −q , A = r iζ , and B = ∂x . The bounded functions q(x) r − ∂ ∂x
and r(x), not necessarily real, are potentials, and ζ is the eigenvalue. It can easily be shown that there is a direct link between the pair (9.10.3a), (9.10.3b) and the original Schrödinger equation. Differentiating (9.10.3b) with respect to x gives ψ2xx = iζψ2x + rx ψ1 + rψ1x ,
(9.10.5)
provided that rx exists. Using (9.10.3ab), equation (9.10.5) can be simplified to obtain ψ2xx = iζψ2x + rx ψ1 + r(−iζψ1 + qψ2 ) 1 = iζψ2x + (rx − iζr)(ψ2x − iζψ2 ) + qrψ2 . r Or equivalently, ψ2xx −
rx rx 2 ψ2x − qr − iζ − ζ ψ = 0. r r
(9.10.6)
When r = −1, this equation reduces to the Schrödinger equation for ψ2 Lψ2 + ζ 2 ψ2 = 0,
(9.10.7)
with q = −u and λ = ζ 2 . Thus the system of equations (9.10.4ab) recovers the scattering equation required for solving the KdV equation. The choice r = −1 turns out to be a degenerate case since, for all other evolution equations, we assume that both q(x) and r(x) decay sufficiently rapidly as |x| → ∞. This ensures the existence of ψ for x ∈ R.
508
9 Solitons and the Inverse Scattering Transform
We close this section by adding the following comments. Both the ZS and AKNS methods are applicable to KdV and NLS equations and to mKdV and SG equations. Although the two methods overlap, the remarkable difference is that the ZS method is described solely in terms of operators, whereas the AKNS method is expressed in terms of the scattering theory.
9.11 Asymptotic Behavior of the Solution of the KdV–Burgers Equation The KdV–Burgers (KdVB) equation ut + 2uux + βuxxx − νuxx = 0,
(9.11.1)
where β and ν(> 0) are constants, combines nonlinearity, linear dispersion and dissipation. This equation describes physical phenomena such as weak shock waves in plasmas, propagation of waves in a liquid-filled elastic tube, and the evolution of liquid–gas mixtures. We consider the asymptotic behavior of the solution of the Cauchy problem for the KdVB equation (9.11.1) with the initial condition u(x, 0) = u0 (x), x ∈ R, as t → ∞. If the initial function u0 (x) decreases sufficiently rapidly at infinity, the uniform asymptotic behavior of this Cauchy problem, as t → ∞, is given by 2 1 x A +O , (9.11.2) u(x, t) ∼ √ exp − 4t t t √ ≥ 0, where the constant A depends explicitly on u0 (x). The with respect to ξ = 2|x| t Cauchy problem for (9.11.1) was investigated by several authors. Bona and Schonbek (1985) proved the existence of traveling wave solutions of (9.11.1) and considered their limiting behavior as a → 0 or b → 0. Jeffrey and Xu (1989) found two exact traveling wave solutions of (9.11.1). On the other hand, Avilov et al. (1987) have investigated the step-decaying problem for the KdVB equation by numerical methods. Their study deals with the asymptotic behavior, as t → ∞, of the solution of the Cauchy problem for (9.11.1) with a sufficiently smooth initial condition u0 (x) tending to ±1 as |x| → ∞. In their study of ion-acoustic waves with Landau damping, Otto and Sudan (1971) and, in his work on weak shock waves and solitons, Ostrovsky (1976) introduced a new nonlinear dispersive equation of the form ∞ μ sgn(x − ξ) (9.11.3) uξ (ξ, t) dξ = 0, ut + u2x + uxxx + λu + (2π)3/2 −∞ |x − ξ|
where λ and μ are nonnegative constants. This equation is now known as the Otto– Sudan–Ostrovsky (OSO) equation, and it describes the nonlinear acoustic effects in the mixture of a gas and liquid bubbles. This equation is a special case of the
9.12 Strongly Dispersive Nonlinear Equations and Compactons
509
Whitham equation (7.8.31), and the symbol K(k) of the operator in (7.8.31) is given by (9.11.4) K(k) = λ + μ |k| − ik 3 . Evidently, when λ = μ = 0, the OSO equation (9.11.3) reduces to the KdV equation. On the other hand, the case μ = 0 and λ > 0 corresponds to the KdV equation with linear dissipation. Several authors including Herman (1990), Kaup and Newell (1978), Karpman and Maslov (1978), and Ko and Kuehl (1980) studied the OSO equation with λ as a small positive parameter and μ = 0. They used the theory of perturbation and the inverse scattering method to solve this equation with only soliton initial conditions and small values of λ. However, the situation with sufficiently general initial perturbations and finite λ seems to be unresolved. In the case μ = 0, the asymptotic behavior, as t → ∞, of the solution of (9.11.3) has an oscillatory character. More detailed asymptotic behavior is available in Chapter 7 of Naumkin and Shishmarev’s book (1994).
9.12 Strongly Dispersive Nonlinear Equations and Compactons Rosenau and Hyman (1993) first discovered a new class of solitary waves with compact support, called compactons. This new class of solutions is governed by a twoparameter family of strongly dispersive nonlinear equations, denoted by K(m, n), ut + a um x + b un xxx = 0, m > 0, 1 < n ≤ 3, (9.12.1a) m n (9.12.1b) ut − a u x + b u xxx = 0, m > 0, 1 < n ≤ 3, for certain values of m and n, where a and b are positive real constants. Thus, compactons are defined as solitons with compact support. In other words, they are solitons with finite wavelength or solitons that are free from exponential trails or wings. Unlike the standard KdV soliton, which narrows as the amplitude (speed) increases, the width of a compacton is independent of the amplitude, but its speed depends on its height. Since dispersion increases with amplitude, at high amplitudes, dispersion is more dominant than in the KdV equation, and hence, it can more effectively counterbalance the effect of nonlinear steepening. Numerous numerical experiments of Rosenau and Hyman (1993) confirmed that, when two or more compactons collide, they undergo a nonlinear elastic interaction according to (9.12.1a) and emerge from the interaction with the original form unchanged. Equation (9.12.1a) with (+a) is called the focusing branch and admits traveling solitary-wave solutions. On the other hand, equation (9.12.1b) with (−a) is referred to as the defocusing branch and admits solitary-wave solutions with cusps or infinite slopes. Thus, equations (9.12.1a), (9.12.1b) represent two nonlinear models with entirely different physical structures. We follow Rosenau and Hyman (1993) to find the solution of K(2, 2) with a = b = 1, that is, the equation (9.12.2) ut + u2 x + u2 xxx = 0.
510
9 Solitons and the Inverse Scattering Transform
Fig. 9.15 A compacton.
We seek a traveling wave solution u = u(ξ), ξ = x − ct of (9.12.2) and integrate the resulting equation twice to obtain the following nonlinear ordinary differential equation: 2 1 1 ∂u c1 + u2 − cu + 2 = c2 , (9.12.3) ∂ξ 4 3 u where c1 and c2 are integration constants. Putting c1 = c2 = 0 leads to the solution 4c ( 3 ) cos2 [ 14 (x − ct)] if |x − ct| ≤ 2π, u(x, t) = (9.12.4) 0 otherwise. This solution is referred to as a compacton and is shown in Figure 9.15. Although the second derivative of the compacton solution is discontinuous at its edges, it represents a solitary wave with compact support because the third derivative acts on u2 which has smooth derivatives everywhere including the edges. It has already been indicated that dispersion increases with amplitude, and is more dominant at higher amplitude than the KdV soliton. Hence, it can more effectively counterbalance the steepening effects of nonlinearity, so the result is a solitary wave with compact support, or compacton. In general, there are three distinct traveling wave solutions of (9.12.3). When c1 = 0, the solutions represent waves that can be described by elliptic functions. When c1 = 0, there exists a singular trajectory that describes a trigonometric wave solution with period 4π and its amplitude depends on the constant c2 . For c2 = 0, the solution u(x, t) is nonnegative and represents a series of compactons. In view of the degeneracy of K(2, 2) at u = 0, these compactons do not interact with each other, and therefore, can be separated. It was also shown by Rosenau and Hyman (1993) that, for a class of general K(m, n) equations, the compacton solution exists only for 1 < n ≤ 3, and the singular dispersion at u = 0 plays a major role in the compactification. The upper limit (n ≤ 3) is necessary for the existence of compacton solutions in the classical sense.
9.12 Strongly Dispersive Nonlinear Equations and Compactons
511
Fig. 9.16 The interaction of three K(2, 2) compactons with speeds c = 2, 1.5, and 1 starting with centers at x = 10, 15, and 40. Rosenau and Hyman (1993).
Based on hundreds of numerical experiments, Rosenau and Hyman (1993) have confirmed that, like solitons, two or more compactons physically interact with each other, and they always remain unchanged after collision except for a slight phase shift. Figure 9.16 exhibits the interaction of three compactons with speeds c = 2, 1.5, and 1 and their identities before and after collision. It was shown by Oron and Rosenau (1989) that K(m, n) type equations arise in the study of nonlinear dispersion in the formation of localized patterns in liquid drops. In their study of a nonlinear model describing new modes of motion of the free surface of a liquid, Ludu and Draayer (1998) demonstrated the existence of localized multiple patterns and nonlinear oscillations which include compactons, solitons, and cnoidal waves as traveling nonaxially symmetric shapes. Subsequently, Ludu et al. (2000) proposed a generalized similarity analysis of nonlinear dispersive equations to find a qualitative description of localized solutions. Their study reveals that compactons fulfill both characteristics of solitons and wavelets with possible new applications to the physics of droplets, bubbles, traveling patterns, fragmentation, fission, and inertial fusions. Dusuel et al. (1998) made an interesting analytical, numerical, and experimental study of physical systems modeled by a nonlinear Klein–Gordon equation with anharmonic coupling, and showed the existence of compactons. In a real physical system, they have also investigated the existence and stability of compactons and kinks consisting of a chain of identical pendulums that are nonlinearly coupled and experience a double-well on-site potential. In general, compacton solutions of K(m, n) equations for any m = n are not yet known. We closely follow the method of solution due to Rosenau and Hyman (1993) and assume the general solution of the K(n, n) equation given by (9.12.1a) in the form
2 (9.12.5) u(x, t) = A sin k(x − ct) n−1 , or in the form
2 u(x, t) = A cos k(x − ct) n−1 ,
where A and k are constants to be determined.
(9.12.6)
512
9 Solitons and the Inverse Scattering Transform
Substituting these solutions into (9.12.1a) and solving the resulting equations for A and k yields ⎧ 1 ⎨ ( 2nc ) n−1 if n is even, a(n+1) (9.12.7ab) A= 1 ⎩ ±( 2nc ) n−1 if n is odd, a(n+1) and
(n − 1) a . k=± 2n b Consequently, the general compacton solutions are: (i) For even n, √
2 u(x, t) = A sin k(x − ct) n−1 H k|x − ct| − 2nπ
and u(x, t) =
2 √ A cos k(x − ct) n−1 H k|x − ct| − nπ ,
where H(|x| − a) = 1, for |x| ≤ a, and zero for |x| > a. (ii) For odd n, √
2 u(x, t) = ± A sin k(x − ct) n−1 H k|x − ct| − 2nπ and
2 √ u(x, t) = ± A cos k(x − ct) n−1 H k|x − ct| − nπ .
(9.12.8)
(9.12.9) (9.12.10)
(9.12.11)
(9.12.12)
Similarly, we seek a solution of the one-dimensional defocusing branch of K(n, n) equation (9.12.1b) in the form
2 u(x, t) = A sinh k(x − ct) n−1 , (9.12.13) or
2 u(x, t) = A cosh k(x − ct) n−1 ,
(9.12.14)
where A and k are constants to be determined. Substituting these solutions in (9.12.1b) and solving the resulting equations for A and k gives the solutions for the sinh-profile, where A and k are given by ⎧ 1 ⎨ ( 2nc ) n−1 if n is even, a(n+1) A= (9.12.15ab) 1 ⎩ ±( 2nc ) n−1 if n is odd, a(n+1) and
(n − 1) k=± 2n
For the cosh-profile, we obtain ⎧ 1 ⎨ −( 2nc ) n−1 , a(n+1) A= 1 ⎩ ±( −2nc ) n−1 a(n+1)
a . b
(9.12.16)
if n is even, (9.12.17ab) if n is odd.
Consequently, the general solutions are given as follows:
9.12 Strongly Dispersive Nonlinear Equations and Compactons
(i) For even n, u(x, t) = and
2 |A|sinh |k|(x − ct) n−1
2 |A|cosh |k|(x − ct) n−1 .
u(x, t) = −
513
(9.12.18) (9.12.19)
(ii) For odd n,
2 u(x, t) = ± |A|sinh |k|(x − ct) n−1 ,
c > 0,
(9.12.20)
and
2 −|A|cosh |k|(x − ct) n−1 ,
u(x, t) = ±
c < 0.
(9.12.21)
With regard to the compactons, it has been shown by Rosenau and Hyman (1993) and Rosenau (1997) that equations K(m, n) for m, n = 2, 3 admit a finite number of local conservation laws. Extensive numerical experiments for m = n = 2, 3 reveal that many of these compactons have a remarkable particle-like robustness that goes far beyond what could be expected from four local conservation laws. Probably, there exist nonlocal conservation laws which play an important role in compacton dynamics. As an example of an application of compactons, we consider a vibration of an anharmonic mass–spring system consisting of N initially equally spaced (h 1) mass points m. The potential part of the associated Hamiltonian is H=
N 1 (yn+1 − yn )Pn (y), h n=1
(9.12.22)
where Pn (y) = N1 αN y N , αN is an anharmonic parameter. For a mixed potential P (y) = 12 α2 y 2 + 13 α3 y 3 , α2 and α3 are anharmonic parameters with small α3 . For the purely quartic potential, Rosenau (1994) obtained the nonlinear Boussinesq 1 2 h , equation of motion in the continuum limit with yx = u, ε = 12 1 2 utt = α3 u + α3 u xx + εα2 uxxxx + 2εα3 q , (9.12.23) 2 xx where
q(ω) = u1−ω uω ux x .
(9.12.24)
Rosenau showed that equation (9.12.23) admits both compacton and usual soliton solutions. For the purely quartic potential in normalized units, the equation of motion becomes (9.12.25) utt = u3 xx + u u2 xx xx . This is clearly a purely cubic nonlinear dispersion equation and fundamentally different from the weakly nonlinear models in that it is nonlinear in the highest-order derivatives, 2u2 uxxxx . Among other features, this equation also admits compacton
514
9 Solitons and the Inverse Scattering Transform
√ solutions in the form 2c cos(x−ct). In addition, it also supports compact breathers of the form u = Q(t)Z(x), where Q(t) satisfies the nonlinear ordinary differential equation in the form (9.12.26) Q (t) + κ2 Q3 (t) = 0, where κ is a separation constant. This equation gives the periodic Jacobi elliptic function solution 1 Q(t) = cn κt, √ . (9.12.27) 2 The function Z(x) satisfies the equation 2 Z Z xx xx + Z 3 xx + κ2 Z = 0, which admits the following compacton solution: √ 8κ cos( 12 x) if |x| ≤ π, Z(x) = 0 otherwise.
(9.12.28)
(9.12.29)
Although many properties of compacton solutions are not yet known, extensive numerical studies indicate that a compacton’s smoothness at the edge is not informative of its stability. These numerical experiments also show that the low-order dispersion is unable to stabilize the compacton, which decomposes immediately into a series of waves. The nonlinear model equation 3 (9.12.30) ut + δu + γu2 + q(ω) + νutxx = 0, 2 x where δ, γ, ω, and ν are constants, admits compacton solutions, and, for 2ω = νγ = 1, it has a bi-Hamiltonian structure. Rosenau (1994) also proved that the infinite sequence of commuting flows generates an integrable, compacton supporting variant of the Harry Dym equation. In summary, the equation governing the motion of a mass–spring system is a prototype of compacton generating equations. With appropriate scalings, the resulting nonlinear model can be applied to study the motion of ion-acoustic waves and the flow of a two-layer liquid. This model also admits compacton solutions. We next discuss physical solid models that are inherently discrete where the lattice spacing represents a fundamental physical parameter. Such discrete models admit compacton solutions, that is, soliton solutions with finite wavelength. Solitontype equations can be derived from such discrete models in which expansions of the wave amplitude and the inverse pulse width normally require a scaling procedure. In other words, the continuum limit approach produces the condition of the slowly varying wave envelope which is consistent with the effect of weak dispersion balanced by a weak nonlinearity. As soon as we deal with compactons instead of typical solitons, the continuum limit approximation can hardly be justified because higher-order derivative terms are numerically small.
9.12 Strongly Dispersive Nonlinear Equations and Compactons
515
We closely follow Kivshar (1993) to consider a one-dimensional lattice model in which each atom interacts with its nearest neighbors by purely anharmonic forces. If un (t) is the nondimensional displacement function of the nth atom from its equilibrium position, and the atoms interact through quartic anharmonic potentials, the equation of motion for the nth atom is given by d2 un = (un+1 − un )3 + (un−1 − un )3 , 2 dt
(9.12.31)
where nondimensional units are employed. In the continuum limit, the particle number is treated as a continuous variable, and the long wavelength excitation of the nonlinear model equation (9.12.31) can be written as (9.12.32) vtt = v 3 xx + · · · , where x = av, a(= 1) is the space of the lattice, and v = (un+1 − un ) is assumed to be a slowly varying function. For short wavelength excitations, the continuum limit approximation can be applied to the wave envelope φn (x, t) defined by the relation un = (−1)n φn (x, t) so that equation (9.12.31) takes the form (9.12.33) φtt + 16φ3 + 6φ φ2 xx + · · · = 0. Using the method of solution due to Rosenau and Hyman (1993), equations (9.12.32) and (9.12.33) can be solved to describe compacton solution properties. However, these nonlinear evolution equations have higher-order dispersive terms that can be neglected because these terms are numerically small for constant-width solutions. Thus, these nonlinear discrete models seem to be natural models for the description of compacton solutions. We assume that φn is independent of time t and then seek standing oscillatory solutions of (9.12.31) in the form un (t) = (−1)n φn F (t).
(9.12.34)
Substituting (9.12.34) into (9.12.31) gives two separable nonlinear equations in the form d2 F + aF 3 = 0, dt2 (φn+1 + φn )3 + (φn−1 + φn )3 = aφn ,
(9.12.35) (9.12.36)
where a is a separation constant. Clearly, equation (9.12.35) admits the Jacobi elliptic function solution in the form F (t) = Acn(ωt, k), √
(9.12.37)
where ω = aA, A is the amplitude, and k = √12 . Assuming a quasi-linear solution with finite wavelength, the method of Rosenau and Hyman (1993) can be used to seek a solution of (9.12.36) in the form
516
9 Solitons and the Inverse Scattering Transform
φn =
cos{θ(n − n0 )}
if |(n − n0 )θ| <
0
otherwise.
π 2,
Substituting (9.12.38) into (9.12.37) gives two relations, 1 π 27 2 θ = , that is, θ = , and a = . tan 2 3 3 4
(9.12.38)
(9.12.39)
Consequently, the general compacton solution of the lattice equation (9.12.31) is given by (−1)n A cos{θ(n − n0 )}cn(ωt, √12 ) if |n − n0 | < 32 , un (t) = (9.12.40) 0 otherwise. If the amplitude of the compacton is taken as an independent parameter, the frequency ω of the compacton can be defined in terms of amplitude A by ω 2 = aA2 .
(9.12.41)
This is identified as the nonlinear dispersion relation. It is evident that the arbitrary parameter n0 represents the center of the compacton (9.12.40) so the n0 = 0 corresponds to the compacton center at the particle site (n = 0). The corresponding compacton pattern is shown in Figure 1(a) given by Kivshar (1993). With only three lattice spacings, the compacton mode involves only three neighboring particles oscillating with opposite phases. At n0 = 0, the solution (9.12.40) can be rewritten as 1 1 1 un (t) = A . . . , 0, − , 1, − , 0, . . . cn ωt, √ . (9.12.42) 2 2 2 This describes the mode pattern through the amplitude of the oscillating particles. On the other hand, when the compacton is centered just between the neighboring particle sites, that is, at n0 = 12 , only two neighboring particles oscillate and the other remain at rest as shown in Figure 1(b) by Kivshar (1993). The mode pattern solution is obtained in the form √ 1 3 √ A(. . . , 0, −1, 1, 0, . . .)cn ωt, , (9.12.43) un (t) = 2 2 √
where 23 A is used as a renormalized amplitude of this mode in order to conserve the total energy. Indeed, solution (9.12.40) describes an infinite family of different localized modes that are characterized by a particular value of n0 ∈ (0, 12 ). Such a compacton solution has been discovered for a chain of particles with quartic interatomic potentials, and can naturally be used to explain the existence of new intrinsic localized modes in anharmonic crystals. Indeed, in their pioneering work, Sievers and Takeno (1988) and Page (1990) discovered these new modes based on the rotating-wave approximation (RWA) in which only a single frequency component
9.12 Strongly Dispersive Nonlinear Equations and Compactons
517
was included in the time dependence. More precisely, the model is described by the equation mw ¨n = k2 (wn+1 +wn−1 −2wn )+k4 (wn+1 −wn )3 +(wn−1 −wn )3 , (9.12.44) where k2 and k4 are the nearest neighbor harmonic and anharmonic force constants. Using the RWA approximation with only the first harmonic contribution, Sievers and Takeno (1988) obtained what is called the odd-parity s-mode with the displacement function 1 1 wn (t) = A . . . , 0, − , 1, − , 0, . . . cos Ωt, (9.12.45) 2 2 where A is the amplitude and Ω is the frequency of the mode above the cutoff fre2 = (4k2 /m) of the linear spectrum band. The solution (9.12.45) is, inquency Ωm deed, the approximate solution of (9.12.44) in the limit as (k4 A2 /k2 ) → ∞. Subsequently, Page (1990) discovered another type of intrinsic localized mode, the evenparity p-mode with the displacement function wn (t) = A(. . . , 0, −1, 1, 0, . . .) cos Ωt.
(9.12.46)
In the above limiting case k4 A2 k2 , the contribution of the nonlinear interaction between particles in the model (9.12.44) is much more significant than that of a linear coupling term, so that this model can be treated as model (9.12.31) for the displacement function un = wn k4 /m which is perturbed by a small linear coupling term. That is why the approximate solutions (9.12.45) and (9.12.46) are very close to the exact solutions (9.12.42) and (9.12.43), respectively. It is pertinent to point out another striking feature of the localized modes in the model (9.12.44) compared to the compacton solution (9.12.40) for the purely anharmonic lattice model described by (9.12.31). Based on a perturbation theory, Sandusky et al. (1992) have demonstrated that the odd-parity s-mode is unstable against certain velocity and displacement perturbations, whereas the even-parity p-mode is absolutely stable against similar perturbations. For positive anharmonicity, both of these modes have amplitudedependent frequencies above the maximum phonon frequency. Furthermore, the unstable odd-parity mode is observed to evolve into several different kinds of moving localized modes. For certain perturbations the odd-parity mode evolves into a mode which smoothly travels from site to site with a constant speed. These traveling modes exist over a wide range of anharmonicity and can become trapped as the anharmonicity increases. As they travel, these modes have a nonconstant phase difference between adjacent relative displacements. Based on the phenomenon of the Peierls–Nabarro potential to the localized mode, Claude et al. (1993) explained this instability of the s-mode. On the other hand, the existence of the exact compacton solution (9.12.40) with arbitrary n0 suggests that the Peierls–Nabarro potential is absent for the compactons and they, therefore, move freely in the lattice provided the interatomic coupling is purely anharmonic in nature. As has been demonstrated by Sievers and Takeno (1988), for sufficiently strong anharmonicity, stable odd-parity localized excitations are possible at any lattice site with a frequency given by
518
9 Solitons and the Inverse Scattering Transform
ω2 ≈
3 27 k2 + k4 A2 , m 16
(9.12.47)
where m is the mass of the atom and A is the amplitude of oscillations of the central atom in the mode pattern (9.12.45). The above analysis also reveals that anharmonicity is fully responsible for the existence of the new intrinsic localized modes in anharmonic quantum crystals at finite temperature. Furthermore, the general compacton solution can describe well two new intrinsic localized modes obtained in the framework of the RWA approximation. Indeed, the compacton (9.12.40) gives the s-mode pattern when it is centered at the particle site, and it reproduces the p-mode pattern when the compacton (9.12.40) is centered in between the nearest particle sites. We close this section by stating higher dimensional focusing branches (+a) and defocusing branches (−a) of K(n, n) equations: (9.12.48) ut + a un x + b un xxx + c un yyy = 0, n n n n ut + a u x + b u xxx + c u yyy + d u zzz = 0, (9.12.49) n n n (9.12.50) ut − a u x + b u xxx + c u yyy = 0, n n n n (9.12.51) ut − a u x + b u xxx + c u yyy + d u zzz = 0, where n > 1 and a > 0. For methods of solution of these equations, the reader is referred to Rosenau and Hyman (1993) and Wazwaz (2002).
9.13 The Camassa–Holm (CH) and Degasperis–Procesi (DP) Nonlinear Model Equations From a physical point of view, it is evident that nonlinearity produces the steepening effects which are counterbalanced by the smoothing effects of dispersion. These effects play a major role in wave peaking and breaking and other physical features of wave phenomena including a variety of weakly singular patterns. In order to understand the major role of these effects, several strongly nonlinear and dispersive models have been developed without a full resolution of the problems, despite over 150 years of progress. Recently, Camassa and Holm (1993) and Camassa et al. (1994) first developed a new, strongly nonlinear, completely integrable model by using asymptotic expansions of the Euler equations for an inviscid incompressible fluid flow in the shallow water regime. Their model is governed by the nonlinear dispersive equation ut − utxx + 3uux = 2ux uxx + uuxxx ,
x ∈ R, t > 0,
(9.13.1)
where u(x, t) is the free surface elevation over a flat rigid bottom. In fact, Fokas and Fuchssteiner (1981) derived (9.13.1) much earlier as an abstract bi-Hamiltonian partial differential equation with infinitely many conservation laws (also see McKean
9.13 The Camassa–Holm (CH) and Degasperis–Procesi (DP)
519
Fig. 9.17 A peakon, or singular solution.
2003). On the other hand, Johnson (2002) gave an alternative derivation of the CH model equation. In fact, Camassa and Holm discovered that their equation is formally integrable in the sense that there exists a Lax pair. Further, the CH model equation has solitary wave solutions which retain their shape and velocity after nonlinear interaction with waves of the same kind, and they are solitons. The most fundamental feature of the CH equation is that it has not only solutions that are global in time but also possesses wave breaking (singular) solutions. In other words, the CH equation has smooth solutions that develop singularities in finite time (or solutions blow-up in finite time). It can be shown that the regular peakon solutions of the CH equation are given by u(x, t) = c exp −|x − ct| , (9.13.2) where c is the velocity of the wave. These are stable solutions (see Constantin and Strauss 2000), and are not classical solutions because they have a peak at their crests and are called weak solutions. These solutions are called peakons, as shown in Figure 9.17. Indeed, the CH solitary waves lead to breaking, the solution remains bounded but its slope becomes unbounded in finite time (see Constantin and Escher 1998a, 1998b, 1998c). It has recently been shown that the CH equation is locally well-posed in the Sobolev space H s (R) for s > 32 , with solutions depending continuously on initial data, and has global conservative solutions in H 1 (R). The peakons have to be understood as weak solutions, as it is suitable to rewrite the CH equation (9.13.1) in the nonlocal conservation law form 1 2 2 −1 2 u 1 − ∂ (ut + uux ) + ∂x u + = 0. (9.13.3) x 2 x Several authors including Dullin et al. (2003) dealt with the traveling wave solutions of a generalized version of the CH equation and found its solutions in an implicit form. Qiao and Zhang (2006) also discussed all possible explicit solutions of the CH equation (9.13.1) with the boundary condition u → A = const. as |x| → ∞. When A = 0, they found the regular peakon solutions, and, when A = 0, both new peaked solitons and a new kind of smooth solitons that can be expressed in terms of trigonometric and hyperbolic functions. Using the traveling wave solution as u(x, t) = η(ξ), ξ = x − ct, with c being the wave velocity, and substituting it in (9.13.1) yields (9.13.4) (η − c) η − η + 2η η − η = 0, where η = dη dξ . The CH equation (9.13.1) has peakon solution of the form u(x, t) = η(ξ) = c exp −|x − ct − ξ0 | ,
520
9 Solitons and the Inverse Scattering Transform
where ξ0 = (x0 − ct0 ), η(ξ0 ) = c, η(±∞) = 0, η (ξ0 −) = c, η (ξ0 +) = −c, η (ξ0 −) and η (ξ0 +) denote the left-hand derivative and the right-hand derivative of η at ξ0 , respectively. It can be verified that the CH equation (9.13.1) has the weak traveling wave solution (9.13.5) u(x, t) = c exp −|x − ct − ξ0 | − sinh |x − ct − ξ0 | , where d and ξ0 are arbitrary real constants. In particular, when A = 0, d = 0, and ξ0 = 0, (9.13.5) reduces to the regular peakon solution (9.13.2) that was originally found by Camassa et al. (1994). When A = 0, both new peaked solitons and a new kind of smooth solitons which can be expressed in terms of trigonometric and hyperbolic functions emerge. On the other hand, Degasperis and Procesi (1999) discovered a new nonlinear dispersive equation, known as the DP equation, in the form ut − uxxt + 4uux = 3ux uxx + uuxxx ,
x ∈ R, t > 0.
(9.13.6)
Both CH and DP equations represent models for the propagation of nonlinear shallow water waves which capture the essential features of wave breaking phenomena. The solution of the DP equation is similar to that the CH equation. In particular, their solutions are singular, leading to wave breaking, that is, they are bounded, but their slopes become unbounded in finite time. Escher et al. (2006) examined weak global solutions and the blow-up feature of the DP equation. The asymptotic analysis and numerical simulations confirmed that the DP equation admits smooth solitons and cusp solitons. Vakhnenko and Parkes (2004) discussed periodic and solitary-wave solutions of the DP equation, and Lenells (2005) investigated the traveling solitary wave solutions of the DP equation. On the other hand, Zhang and Qiao (2007) studied cusp soliton and smooth soliton solutions of the DP equation (9.13.6) with the non-homogeneous boundary conditions u(x, t) → A = 0 as |x| → ∞. They also found by direct calculation that the DP equation (9.13.6) admits a new stationary cusp soliton solution in the form
1 1,1 , (9.13.7) u(x, t) = 1 − exp −2|x| 2 ∈ Wloc for any x = 0. If u ∈ H 1 , the DP equation (9.13.6) can equivalently be expressed into the following nonlocal conservation law form −1 3 2 u = 0. L(u) = ut + uux + ∂x 1 − ∂x2 (9.13.8) 2 However, if u ∈ / H 1 , (9.13.8) is no longer equivalent to the DP equation (9.13.6). The cusp soliton solution does not satisfy (9.13.8) because L(u) = sgn x exp(−|x|). This means that it is impossible to find the cusp soliton solution of the DP equation (9.13.6) through its nonlocal conservation form (9.13.8).
9.13 The Camassa–Holm (CH) and Degasperis–Procesi (DP)
521
The DP equation has not only peakon solitons (9.13.2), but also a shock peakon solution in the form 1 (9.13.9) u(x, t) = − sgn(x)e−|x| . (t + k) This represents a significantly different solution from that of the CH equation. The shock-peakon solutions (see Lundmark 2007) can be observed by substituting (x, t) → (εx, εt) to (9.13.6) and letting ε → 0 so that it yields the inviscid derivative Burgers equation as (ut + uux )xx = 0,
(9.13.10)
which produces shock waves. On the other hand, the celebrated KdV equation is an integrable model for nonlinear shallow water waves and the dispersive effect incorporated into the KdV model prevents breaking. In contrast to the KdV model, the CH and DP model equations capture the essential features of wave breaking. But they do not shed light on the breaking process or what happens after breaking which are the most fundamental features of water waves for which there appears to be no satisfactory mathematical, physical, or computational theory for a long period of time. In recent years, Holm and Staley (2003) investigated the following family of (1 + 1)-dimensional nonlinear viscous equation for the fluid velocity u(x, t) in the form (9.13.11) mt + umx + bux m = νmxx , where u = g ∗ m is the convolution defined by g(x − ξ)m(ξ) dξ, u(x) =
(9.13.12)
R
which relates velocity u to momentum density m with the kernel g(x) over R. The kernel g(x) is chosen to be the Green’s function for the Helmholtz operator (1 − ∂x2 ) on the line, that is, g(x) = 12 exp(−|x|). This means that m = (u − uxx ). The family of equations (9.13.11) is characterized by the kernel g(x) and the real nondimensional parameter b which is the ratio of the stretching term (bux m) to the convective term (umx ) in (9.13.11). The parameter b is also the number of covariant dimensions associated with m, and ν (> 0) is the viscosity coefficient associated with m. It is to be noted that the kernel g(x) determines the traveling wave shape and length scale for equation (9.13.11), whereas the constant b represents a balance (or bifurcation) parameter for the nonlinear solution. The quadrative terms in (9.13.11) describe the balance in fluid convection between nonlinear transport and amplification due to bdimensional stretching term. In a recent work on soliton dynamics, it is shown that equation (9.13.11) for ν = 0 and b = −1 is included in the family of shallow water equations at quadratic order accuracy that are asymptotically equivalent under Kodama transformations (see Dullin et al. 2003). In the absence of viscosity (ν = 0), equation (9.13.11) reduces to (1 + 1)dimensional b-family equations in the form
522
9 Solitons and the Inverse Scattering Transform
mt + umx + bmux = 0,
u = g ∗ m.
(9.13.13)
Using the method of asymptotic integrability, Degasperis and Procesi (1999) showed that equation (9.13.13) cannot be completely integrable unless b = 2 or b = 3. When b = 2, equation (9.13.13) becomes the CH equation (9.13.1). When b = 3, equation (9.13.13) reduces to the DP equation (9.13.6). It is shown by Dullin et al. (2003) that the DP equation can be derived from the shallow water elevation equation by an appropriate Kodama transformation. Lundmark and Szmigielski (2003) used the universe scattering method for finding n-peakon solutions of the DP equation, and Holm and Staley (2003) used numerical method to study stability of solitons and peakons. Recently, Gui et al. (2008) investigated the local well-posedness for the peakon b-family of equation (9.13.13) that includes both the CH and the DP equations as special cases and found the precise blow-up feature of strong solutions of (9.13.13) with certain initial conditions. Among the singular entities, the peakon, a soliton with a finite discontinuity in gradients at its crests, is perhaps the weakest nonanalytic solution ever observed. The peakon solutions have been known for some time (see Fuchssteiner 1981). Camassa and Holm (1993) proved the integrability of equation 1 (9.13.14) ut − uxxt = bux + 3uux − uuxx + u2x . 2 x Even if b = 0 in (9.13.14), it admits peakon solutions u(x, t) = c exp(−|x + ct|) which are obtained as a solution of the equation (c − u) u2ξ − u2 = 0, ξ = x + ct. (9.13.15) If b = 0, equation (9.13.14) yields soliton solutions which are analytic functions. Another nonlinear model due to Camassa and Holm (1993) is described by the equation (u ± uxx )t = bux +
1 2 u ± u2x (u ± uxx ) x . 2
(9.13.16)
This equation also admits peakon solutions if b > 0. When b < 0, ordinary solitons emerge, but if b = 0, no solitons are possible. For a detailed discussion of this equation (9.13.16) and its peakon solutions, the reader is referred to Rosenau (1997). It is important to note that the singular solutions of the Camassa–Holm equation look very similar to Stokes’ wave of extreme height, but they only appear in the rest frame of reference. This equation is not Galilean invariant. Another generalized version of the Camassa–Holm equation is ut − uxxt + 3uux + 2κux = 2ux uxx + uuxxx ,
(9.13.17)
where κ is a real constant. It has many conservation laws and it has peakon solution when κ = 0; and its solitary wave solution is stable for κ > 0. The local well-posedness, global existence, blow-up phenomena, and the well-posedness of
9.13 The Camassa–Holm (CH) and Degasperis–Procesi (DP)
523
global weak solutions of (9.13.17) have been discussed by many authors including Constantin and Escher (1998a, 1998b, 1998c), and Constantin and McKean (1999). Bressan and Constantin (2007) obtained the sharpest results for equation (9.13.17) for the global existence and blow-up solutions. Parker (2004, 2005a, 2005b) and Ohta et al. (2008) constructed soliton solutions of equation (9.13.17). For κ = 0, the soliton solution of (9.13.17) for the initial value problem is represented graphically by Camassa and Holm (1993). Indeed, the solution for the Gaussian initial condition breaks up into a series of ordered solitons as time progresses. Thus, the soliton train eventually wraps around the periodic domain so that the leading solitons overtake the slower emergent solitons from behind after interaction that causes phase shifts. Hence, the CH equation (9.13.17) represents a model of unidirectional propagation of shallow water waves over a flat bottom. It also arises as a model equation for the axially symmetric waves in a hyperelastic rod. The analysis of the CH equation has provided a challenge and a source of inspiration for many new developments in nonlinear water waves, integrable systems, asymptotics, geometry and Lie groups. Camassa and Holm (1993) discovered the peakon solitary traveling wave solution of (9.13.17) for nonlinear shallow water waves in the form |x − ct| , (9.13.18) u(x, t) = c exp − α where the fluid velocity u(x, t) is a function of position x in R and time t. The peakon solitary wave moves with a velocity equal to its maximum height, at which it has a sharp peak with jump in derivative. So, peakons can be obtained after solving the initial value problem for a partial differential equation derived by an asymptotic expansion of the Euler equations using the small parameters of shallow water wave dynamics. In fact, peakons are nonanalytic solitons that are the superposition of N soliton solutions as u(x, t) =
N 1 pn (t) exp −x − qn (t)/α , 2 n=1
(9.13.19)
where the discrete sets {pn (t)} and {qn (t)} of peakon parameters satisfy the canonical Hamiltonian equations. They also arise in the absense of linear dispersion (κ = 0) in (9.13.17). Each term in (9.13.19) is a soliton with a sharp peak at its maximum. Expressed using its momentum m = (1 − α2 ∂x2 )u, the peakon velocity solution (9.13.19) of dispersionless CH equation becomes a sum of delta functions, supported on a set of points traveling on the real axis. In other words, the peakon velocity (9.13.19) implies N pn (t)δ x − qn (t) , (9.13.20) m(x, t) = α n=1
due to the fact that (1−α2 ∂x2 ) exp(− |x| α )
= 2αδ(x). These solutions satisfy (9.13.17) for the zero dispersion parameter (κ = 0).
524
9 Solitons and the Inverse Scattering Transform
Substituting the peakon solution (9.13.19) and (9.13.20) into the dispersionless CH equation (9.13.21) mt + umx + 2mux = 0, m = u − α2 uxx , gives the celebrated Hamilton’s canonical equations for the discrete set of peakon parameters pn (t) and qn (t) as q˙n (t) =
∂HN ∂pn
and
p˙n (t) = −
∂HN , ∂qn
(9.13.22)
where the Hamiltonian HN is given by HN =
N 1 |qn − qm | , pn pm exp − 4 n,m=1 α
(9.13.23)
n, m = 1, 2, . . . , N . In their recent work of weakly nonlinear analysis combined with the variational principle for Euler equations, Holm and Staley (2003) made a two-dimensional generalization of the dispersionless CH equation (9.13.21). This generalization is known as the Euler–Poincaré (EP) equation for the Lagrangian consisting of the kinetic energy 1 2 K= |u| + α2 (div u)2 dx dy, (9.13.24) 2 where u = (u, v) is a two-dimensional fluid velocity vector. They have shown that evolution generated by the kinetic energy in Hamilton’s principle results in geodesic motion with respect to the velocity norm u which is provided by the Lagrangian of the kinetic energy. The EP equation produced by any choice of kinetic energy norm without imposing incompressibility is called “EPDiff ” for the Euler–Poincaré equation for geodesic motion on the diffeomorphisms. Thus, the EPDiff equation given by Holm and Staley (2003) is ∂ + u · ∇ m + ∇uT · m + m(div u) = 0, (9.13.25) ∂t 1 2 where m = δK δu is the momentum density, K = 2 u is the kinetic energy which defines a norm in the fluid velocity u. Clearly, this equation has no contribution from either the pressure, or the potential energy. However, this equation conserves the velocity norm u given by the kinetic energy. In fact, its evolution describes geodesic motion on the diffeomorphisms with respect to this norm. There is an alternative way of expressing the EPDiff equation (9.13.25) in either two or three dimensions as
∂m − u × curl m + ∇(u · m) + m(div u) = 0. ∂t
(9.13.26)
Remarkably, all three differential operators—gradient, divergence, and curl—are present in this form of EPDiff equation. For the kinetic energy Lagrangian K given
9.13 The Camassa–Holm (CH) and Degasperis–Procesi (DP)
525
by (9.13.24) which is a norm for an irrotational flow (curl u = 0), (9.13.25) is the EPDiff equation with the momentum m = δK/δu = u − α2 ∇(div u). It is important to point out that the EPDiff equation (9.13.25) has many other interpretations beyond applications in fluid flows. However, the stability problem for EPDiff singular momentum solutions has not yet been solved, even through the stability of the peakon in one dimension was proved by Constantin and Strauss (2000). On the other hand, Degasperis and Procesi (1999) derived another generalized form of the shallow water model equation ut − uxxt + 2κux + 4uux = 3ux uxx + uuxxx ,
(9.13.27)
Like (9.13.17), Degasperis et al. (2002) proved that (9.13.27) is also completely integrable and has the Lax pair formulation. It admits an infinite number of conservation laws and its solitary waves interact like solitons (see Matsuno 2005a, 2005b). Recently, Constantin and Lannes (2009) derived a generalized version of equations (9.13.17) and (9.13.18) in the form 3 ut − ux + εuux + δ(αuxxx + βuxxt ) = εδ(aux uxx + buuxxx ), (9.13.28) 2 where ε and δ are defined by (9.3.2ab), α, β, a, and b satisfy certain conditions. They also proved the large time well-posedness on a time scale O(|ε|−1 ) provided the initial value u0 belongs to H s for s > 52 . They also discussed the wave breaking phenomena. Using suitable transformations, equation (9.13.28) can be transformed into the form ut − uxxt + 2κux + αuux = aux uxx + buuxxx ,
(9.13.29)
where κ, α, a, and b are constants. Obviously, (9.13.29) is a generalization of both equations (9.13.17) and (9.13.27). The major difference between (9.13.29) for the case a = 2b and the CH equation (9.13.17) is that equation (9.13.29) does not have the conservation law 2 I= u + u2x dx (9.13.30) R
which plays a major in the investigation of the CH equation (9.13.17). Recently, Lai and Wu (2011) investigated local existence and uniqueness of solution for the generalized equation (9.13.29). They also proved the local well-posedness of solutions of (9.13.29) in the Sobolev space H s (R) for s > 32 . We close this discussion by simply stating two other strongly nonlinear models, the nonlinear shallow water model, and the Harry Dym equation (see Kruskal 1975, Leo et al. 1983), 1 (9.13.31) ut = u− 2 xxx , which arises as a generalization of the class of isospectral flows of the Schrödinger operator. They lead to essentially new scattering problems, which have not yet been fully explored. There is another version of Harry Dym equation, 1 1 ∂ , (9.13.32) ut = D3 − 4D u− 2 , D ≡ 2 ∂x with an extra transport term.
526
9 Solitons and the Inverse Scattering Transform
A recent literature search also suggests that there is a class of hierarchies containing the KdV hierarchy that are obtained from the isospectral problem D2 + k 2 u(x) − q(x).
(9.13.33)
These include the KdV flows, the negative KdV flows, the Harry Dym equation, and the Camassa–Holm equation (see Qiao 2003). Almost all studies of steady and unsteady water waves are based on the assumption that flows are irrotational (that is, curl u = 0). In general, this is a good assumption. However, vorticity is generated at the free surface by wind stress or current, or by friction at the rigid bottom boundary. In fact, a very thin vorticity (or shear) layer is produced at the free surface or at the bottom boundary. In recent years, considerable attention has been given to study water waves by assuming for simplicity that vorticity is constant in the fluid flows. Simmen and Saffman (1985) first investigated numerical solutions for periodic waves in deep water with constant vorticity. Their study showed that the waves have either a limiting configuration with a 120◦ angle at the crests or a trapped bubble at their troughs. Following Simmen and Saffman (1985), we consider a two-dimensional periodic waves in an inviscid incompressible fluid of infinite depth. The flow is assumed to be rotational and described by a constant vorticity, Ω. We take a frame of reference with the x-axis along the mean water level and in which the flow is steady with gravity is in the negative y-axis. We assume that the steady flow is symmetric with respect to the y-axis. The flow can be described in terms of a stream function ψ(x, y) satisfying ∇2 ψ = Ω
(9.13.34)
in the flow domain. We reduce the problem to one governed by the Laplace equation by subtracting a particular solution of (9.13.34). Thus, if we write ψ=Ψ+
Ω 2 y − cy, 2
(9.13.35)
then ∇2 Ψ = 0
(9.13.36)
with the requirement that Ψ → 0 as y → −∞. This defines the quantity c in (9.13.35) uniquely and c is referred to as the wave velocity. In terms of the dimensionless variables by choosing λ as the unit length and c as unit velocity, the dimensionless equation (9.13.35) becomes ψ=Ψ+
ω 2 y − y, 2
where ω is the dimensionless vorticity defined by Ωλ . ω= c
(9.13.37)
(9.13.38)
9.13 The Camassa–Holm (CH) and Degasperis–Procesi (DP)
527
The function w(z) = u − iv = Ψy + iΨx is an analytic function of z = x + iy, and the fluid velocity vector is (u + ωy − 1, v). The function w(z) vanishes at infinity. Vanden-Broeck’s (1996a, 1996b) numerical analysis first confirmed the results of Simmen and Saffman (1985). He then found that there are solution branches that bifurcate from the uniform shear flow (u, v) = (0, 0). As we move along the solution branches, the waves ultimately attain limiting configurations with a 120◦ angle at the crests or a trapped bubble at their troughs. Furthermore, new families of solutions can also be constructed. Several authors including Teles da Silva and Peregrine (1998), Pullin and Grimshaw (1988), and Vanden-Broeck (1994, 1995) obtained numerical solutions for solitary waves with constant vorticity. They formulated the problem in terms of a stream function ψ such that (9.13.39) ∇2 ψ = −ω, where the flow is rotational and characterized by a constant vorticity Ω so that ω = (Ωh/c) is the nondimensional vorticity, h is the depth of the fluid, and c is the wave velocity. These authors reduced the problem to the Laplace equation ∇2 Ψ = 0 by subtracting a particular solution of (9.13.39), that is, ψ = Ψ − { ω2 y 2 − (1 + ω)y}. Thus, w(z) = u − iv = Ψy + iΨx is an analytic function of z = x + iy, where the fluid velocity vector is {u − ω(y − 1) + 1, v}. They require the kinematic condition that v = 0 on the bottom by reflecting the flow in the bottom. The function w(z) vanishes at infinity. This problem is then numerically investigated using the flow variables nondimensional by choosing h as the unit length and c as the unit velocity. The flow is then described by the above parameter ω, and two other parameters η gh and a = , (9.13.40) G= c2 h where η is the elevation of the wave crests above the level of the free surface. Clearly, G is the nondimensional gravitational parameter, and a is the amplitude parameter. When ω = 0, the flow becomes irrotational. The above authors numerically examined solutions for various values of the parameters ω, G, and a. Their solitary wave solutions are solution branches that bifurcate from the trivial solution (u = v = 0 and a = 0) at the critical values G = 1 + ω which is obtained by Benjamin (1962). He studied asymptotic solutions for solitary waves of small amplitude. His analysis showed that, for each value of ω, there is a one-parameter family of solutions that bifurcates from the uniform shear flow at the critical value G = 1 + ω. He also found an asymptotic solution for small values of the parameter a and obtained the following relation between G, ω, and a: ω2 . (9.13.41) G=1+ω−a 1+ω+ 3 Vanden-Broeck (1994, 1995) and others generalized Benjamin’s results for waves of finite amplitude. They obtained some solution branches which are not associated with Benjamin’s solutions and have shown that their numerical values agree with
528
9 Solitons and the Inverse Scattering Transform
(9.13.41) as a → 0. As they progress along the solution branches, there exists a critical value ωc ≈ −0.32 of ω so that different limiting configurations occur for ω > ωc and ω < ωc . For ω < ωc , they found that the solution branches which bifurcate from a uniform shear flow at the critical value G = 1 + ω do not have limiting configuration with 120◦ angle at the crests. In the limit as G → 0, the solution branches eventually tend to closed regions of constant vorticity in contact with the bottom boundary if G is allowed to become negative. For ω > ωc , the solution branches that bifurcate from a uniform shear flow at the critical values G = 1 + ω have limiting configuration with 120◦ angle at their crests (see Teles Da Silva and Peregrine 1988 and Pullin and Grimshaw 1988). As a concluding remark, it is necessary to point out that the boundary integral equation method is used by Vanden-Broeck (1996a, 1996b) to compute periodic waves with constant vorticity and to find the solution in the limit as G → 0. This computational results reveal that there are solution branches that tend to configurations with a closed region of fluid in rigid body rotation as G → 0. Further, there are solitary waves with circular closed regions at their crests. Since the solitary waves can be considered as the limit of periodic waves as (λ/h) → ∞, Vanden-Broeck’s (1996a, 1996b) numerical analysis strongly suggest that configurations with circular regions at the wave crests are a general feature of waves in water of finite depth.
9.14 Exercises 1. Show that (xu + 3tu2 ) is a conserved density for the KdV equation (9.7.1). 2. Find three conservation laws for the mKdV equation (Miura et al. 1968). ut − 6u2 ux + uxxx = 0,
x ∈ R,
which involve u, u2 , and u4 , respectively. 3. Show that the three conservation laws for the BBM equation ut − uux − uxxt = 0 are
1 2 (i) ut − uxt + u = 0, 2 x 1 2 1 3 2 u + ux t − uuxt + u (ii) = 0, 2 3 x
and (iii)
1 3 u 3
+ t
u2t
−
u2xt
1 − u uxt − u4 4
2
4. Show that the mKdV equation ut + 6u2 ux + uxxx = 0
= 0. x
9.14 Exercises
529
is invariant under the transformation x ˜ = ax,
t˜ = a3 t,
u ˜ = a−1 u (a = 0).
Hence introduce u(x, t) = t−1/3 f (xt−1/3 ) to show that f (ξ) satisfies the nonlinear ordinary differential equation 1 f − ξf + 2f 3 = 0, 3
ξ = xt−1/3 ,
provided that f → 0 at infinity. 5. Investigate a similarity solution of the cylindrical KdV equation 1 1 2ut + u − 3uux + uxxx = 0 t 3 in the form u(x, t) = − 13 ( t22 ) 3 f {x(2t)−1/3 }. Show that v(ξ) satisfies the equation v − ξv + v 3 = 0, 1
where f = v2 and ξ = x(2t)−1/3 . 6. Show that the NLS equation iut + uxx + γu|u|2 = 0, where γ is a real constant, is invariant under each of the group transformations ˜ = a−1 u (a = 0), (i) x ˜ = ax, t˜ = a2 t, u (ii) x ˜ = x + a, t˜ = t, u ˜ = u. Show that a similarity solution of the form u(x, t) = tp f (ξ), ξ = xtq exists for suitable values of p and q. Hence, find the equation for f (ξ). 7. Show that the Kadomtsev–Petviashvili equation (Freeman 1980) (ut − 6uux + uxxx )x + 3uyy = 0 has the soliton solution
1 2 2 1 u(x, y, t) = − k sech (kx + y − ωt) , 2 2
where ω = k3 + (3 2 /k). 8. Apply the inverse scattering transform method to solve the initial-value problem for the KdV equation (9.7.1) with u(x, 0) = 2 sech2 x. Hence, determine the time development of the solution from the above initial data by numerical integration with a program given by Crandall (1991).
530
9 Solitons and the Inverse Scattering Transform
9. Use the Bäcklund transformations vx = u − v 2
and vt = −uxx + 2(uvx + ux v)
to show that u satisfies the KdV equation ut − 6uux + uxxx = 0, and that v satisfies the mKdV equation vt − 6v 2 vx + vxxx = 0. 10. Apply the Bäcklund transformations vx = −
1 (uv) 2ν
and
vt =
v 2 u − 2vux 4ν
to show that u and v satisfy the Burgers and diffusion equations ut + uux = νuxx
and
vt = νuxx .
11. Show that the Kadomtsev and Petviashivili (1970) equation (ut − 6uux + uxxx )x + 3uyy = 0 can be derived by selecting the Lax pair ∂2 ∂ L=− 2 + +u ∂x ∂y
and
∂ ∂3 M = −4 3 + 6u + 3ux + 3 ∂x ∂x
x
uy dx.
12. Show that the operator M for the KdV equation is M = −4
∂3 ∂u ∂ +3 . + 6u 3 ∂x ∂x ∂x
13. Using Mathematica, or otherwise, show that the Lax equation (9.9.6) with ∂ ∂3 +u , ∂x3 ∂x ∂5 ∂3 ∂u ∂ ∂2u ∂ M = 9 5 + 15u 3 + 15 + 5u2 + 10 2 ∂x ∂x ∂x ∂x ∂x ∂x L=
reduces to the equation ∂5u ∂ 3u ∂u ∂ 2 u ∂u ∂u = · + 5u 3 + 5 + 5u2 . 5 ∂t ∂x ∂x ∂x ∂x2 ∂x For q = ar − 1, and matrices A and B given by (9.9.41) and (9.9.45), show that (9.9.40) leads to the nonlinear equation rt + rxxx − 6rrx + 6ar2 rx = 0.
9.14 Exercises
531
14. If the operator M in the Lax equation (9.9.15) is n
M = −αD2n+1 +
Um D2m−1 + D2m−1 Um + A(t),
m=1
where α is a constant, um = um (x, t), and A(t) is an arbitrary function, show that the Lax equation reduces to the KdV equation (9.9.21) for n = 1, and, for n = 2, it becomes the fifth-order KdV equation ut + 30u2 (Du) − 20(Du) D2 u − 10u D3 u + D 5 u = 0. 15. If the phase velocity of nonlinear water waves is c(k) =
12 g 1 tanh kh = gh 1 − k2 h2 + o k 2 h2 as kh → 0, k 6
show that these waves are described by the Whitham equation ∞ ∂u dξ = 0, ut + uux + K(x − ξ) ∂ξ −∞ where the kernel K is determined from the linear theory as the Fourier transform of c(k), ∞ 1 c(k) exp(ikx) dk. K(x) = 2π −∞ 16. Show that the Boussinesq equation utt − uxx + 3 u2 xx − uxxxx = 0 has the ∞following conservation laws: (a) −∞ u dx = const. (conservation of mass), ∞ (b) −∞ ut dx = const. (conservation of momentum). 17. Show that the solitary wave solution of the Boussinesq equation utt − uxx + 3 u2 xx − uxxxx = 0 is
u(x, t) = a sech2 b(x − ct) + d ,
for suitable relations between the constants a, b, c, and d. Verify that the Boussinesq wave propagates in either direction. 18. Show that the KP equation (ut − 6uux + uxxxx )x + 3uyy = 0 has the solitary wave solution u(x, y, t) = a sech2 (kx + ly − ωt + α) for suitable relations between the constants a, k, l, ω, and α.
532
9 Solitons and the Inverse Scattering Transform
19. Show that the KdV equation 1 2ut + 3uux + uxxx = 0 3 has the ∞following conservation laws (see Johnson 1997): (a) −∞ u dx = const. (conservation of mass), ∞ (b) −∞ (u3 − 13 u2x ) dx = const. (conservation of energy), ∞ (c) −∞ u2 dx = const., ∞ 45 4 (d) −∞ ( 4 u − 15uu2x + u2xx ) dx = const. 20. (a) Show that the CH equation (9.13.1) can be expressed in the nonlocal conservation law form (see Bressan and Constantin 2007) 2 u 1 2 2 −1 2 u + ux = 0. + ∂ x 1 − ∂x ut + 2 x 2 (b) Verify that the equation in Exercise 20(a) is equivalent to 2 u + Px = 0, ut + 2 x where P is defined as a convolution P =
1 1 exp(−|x|) ∗ u2 + u2x . 2 2
(c) Differentiate the equation in Exercise 20(b) with respect to x to obtain 1 2 2 2 uxt + uuxx + ux − u + ux + P = 0. 2 (d) Derive two conservation laws with source term 2 3 u u + uP + = ux P, 2 3 x 2 t 2 u3 uux ux − + = −ux P. 2 t 2 3 x Hint: Multiply the equation in Exercise 20(b) by u and the equation in Exercise 20(c) by ux to derive the two conservation law. (e) Show that the total energy for regular solutions is 2 E(t) = u + u2x dx, R
is constant in time.
9.14 Exercises
533
21. (a) Consider the Cauchy problem for the dispersive DP equation with m = u − uxx (see Guo 2009): mt + umx + 3ux m + γmx = 0,
x ∈ R, t > 0,
m(x, 0) = u0 (x) − u0xx (x), where γ is a real constant. If p(x) =
1 − ∂x2
−1
1 2
exp(−|x|), x ∈ R, show that
f =p∗f
for all f ∈ L2 (R),
and p ∗ m = u. Using these results, verify that the Cauchy problem can be written in the form 3 2 u + γux = 0, x ∈ R, t > 0 ut + uux + ∂x p ∗ 2 u(x, 0) = u0 (x). (b) Show that the Cauchy problem in Exercise 21(a) with the initial data u0 (x) has a unique global solution u(x, t) which is the sum of a peakon and an antipeakon solution in the form u(x, t) = p1 (t) exp −x − q1 (t) + p2 (t) exp −x − q2 (t) , 1,∞ for some pi , qi ∈ Wloc (R), i = 1, 2. 22. Show that u(x, t) = (c − γ) exp(−|x − ct|)(c = γ) is a global weak solution of the dispersive DP equation given in Exercise 21(a).
10 The Nonlinear Schrödinger Equation and Solitary Waves
. . . Schrödinger and I both had a very strong appreciation of mathematical beauty, and this appreciation of mathematical beauty dominated all our work. It was a sort of act of faith with us that any equations which describe fundamental laws of Nature must have great mathematical beauty in them. It was like a religion with us. It was a very profitable religion to hold, and can be considered the basis of much of our success. Paul Dirac . . . the progress of physics will to a large extent depend on the progress of nonlinear mathematics, of methods to solve nonlinear equations . . . and therefore we can learn by comparing different nonlinear problems. Werner Heisenberg
10.1 Introduction It has already been indicated in Section 2.3 that the nonlinear Schrödinger (NLS) equation arises in a wide variety of physical problems in fluid mechanics, plasma physics, and nonlinear optics. The most common applications of the NLS equation include self-focusing of beams in nonlinear optics, modeling of propagation of electromagnetic pulses in nonlinear optical fibers which act as wave guides, and stability of Stokes waves in water. Some formal derivations of the NLS equation have been obtained by several methods which include the multiple scales expansions, the asymptotic method, Whitham’s (1965a, 1965b) averaged variational equations, and Phillips’ (1981) resonant interaction equations. Zakharov and Shabat (1972) developed an ingenious inverse scattering method to show that the NLS equation is completely integrable. The NLS equation is of great importance in adding to our fundamental knowledge of the general theory of nonlinear dispersive waves. It is now well known that the Korteweg–de Vries equation and the nonlinear Schrödinger equation are the lowest-order nontrivial consequences of a perturbation L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_10,
536
10 The Nonlinear Schrödinger Equation and Solitary Waves
approximation for weakly and strongly nonlinear dispersive wave systems, respectively. This chapter is devoted to the nonlinear Schrödinger equation and its solitary wave solutions. Several properties of the nonlinear Schrödinger equation including envelope solitons and recurrence phenomenon are discussed. Attention is given to conservation laws for the nonlinear Schrödinger equation. Some examples of applications to fluid dynamics, plasma physics, and nonlinear optics are included. Based on an ingenious method of Zakharov and Shabat (1972), the inverse scattering transform for the nonlinear Schrödinger equation is described in some detail.
10.2 The One-Dimensional Linear Schrödinger Equation We consider the following Fourier integral representation of a quasi-monochromatic plane wave solution: ∞ F (k) exp i kx − ω(k)t dk, (10.2.1) φ(x, t) = −∞
where the spectrum function F (k) is determined from the given initial or boundary conditions and has the property F (−k) = F ∗ (k), and ω = ω(k) is the dispersion relation. We assume that the initial wave is slowly modulated as it propagates in a dispersive medium. For such a quasi-monochromatic wave, most of the energy is confined in the neighborhood of a specified wavenumber k = k0 , so that the spectrum function F (k) has a sharp peak around k0 with a narrow wavenumber width (k − k0 ) = δk = O(ε), and the dispersion relation ω(k) can be expanded about k0 as follows: 1 ω = ω0 + (δk)ω0 + (δk)2 ω0 + · · · , 2 where ω0 = ω(k0 ), ω0 = ω (k0 ), and ω0 = ω (k0 ). Substituting (10.2.2) in (10.2.1) gives φ(x, t) = ψ(x, t) exp i{k0 x − ω0 t} + c.c.,
(10.2.2)
(10.2.3)
where c.c. stands for the complex conjugate and ψ(x, t) is the complex amplitude defined by ∞ 1 2 ψ(x, t) = F (k0 + δk) exp i(x − ω0 t)δk − iω0 (δk) t d(δk), (10.2.4) 2 0 where it has been assumed that ω(−k) = −ω(k). Since (10.2.4) depends on (x − ω0 t)δk and (δk)2 t where δk = O(ε) is small, the amplitude ψ(x, t) is a slowly varying function of x∗ = (x − ω0 t) and t. We next introduce slow variables ξ and τ defined by ξ = ε(x − ω0 t) and τ = ε2 t, (10.2.5ab) so that the wave field assumes the form
10.3 The Nonlinear Schrödinger Equation and Solitary Waves
φ(x, t) = A(ξ, τ ) exp i(k0 x − ω0 t) + c.c.,
537
(10.2.6)
where the modulated wave amplitude is given by
1 2 A(ξ, t) = F (k0 + δk) exp i δk(x − ω0 t) − ω0 (δk) t d(δk) 2
1 = F (k0 + ε) exp i ξ − ω0 τ d(ε). (10.2.7) 2 In the expression for the wave field (10.2.6), k0 , and hence, ω0 = ω(k0 ) are chosen to be constants, and all slow variations of the wavetrain are included in A(ξ, τ ). A simple computation gives 1 Aτ = − iω0 A 2
and
Aξξ = −A
(10.2.8ab)
so that the modulated wave amplitude A(ξ, τ ) satisfies the linear Schrödinger equation 1 (10.2.9) iAτ + ω0 Aξξ = 0. 2
10.3 The Nonlinear Schrödinger Equation and Solitary Waves We show below that the nonlinear modulation of a quasi-monochromatic wave is described by the nonlinear Schrödinger equation. To take into account the nonlinearity and the modulation in the far-field approximation, the wavenumber k and frequency ∂ ∂ ω in the linear dispersion relation are replaced by k − i ∂x and ω + i ∂t , respectively. It is convenient to use the nonlinear dispersion relation in the form ∂ ∂ (10.3.1) D k − i , ω + i , |A|2 A = 0. ∂x ∂t We consider the case of a weak nonlinearity and a slow variation of the amplitude, and hence, the amplitude A is assumed to be a slowly varying function of space ∂ ∂ and time. We next expand (10.3.1) with respect to |A|2 , −i ∂x , and i ∂t to obtain ∂ ∂ A − Dω D(k, ω, 0) − i Dk ∂x ∂t ∂2 ∂2 ∂2 1 − Dkk 2 − 2Dkω + Dωω 2 A 2 ∂x ∂x∂t ∂t ∂D + |A|2 A = 0, (10.3.2) ∂|A|2 where the first term D(k, ω, 0) = 0 is due to the linear dispersion equation. Introducing the transformation x∗ = x − Cg t, t∗ = t, assuming that A = O(ε), ∂ ∂ 2 3 ∂x∗ = O(ε), and ∂t∗ = O(ε ), retaining all terms up to O(ε ), and dropping the asterisks, we find that
538
10 The Nonlinear Schrödinger Equation and Solitary Waves
∂A ∂2A + p 2 + q|A|2 A = 0, ∂t ∂x 1 1 dCg ∂D , q= p= , 2 dk Dω ∂|A|2 i
where
(10.3.3)
(10.3.4)
and the following results: Cg = −
Dω , Dk
dCg = Dkk + 2Cg Dωk + Cg2 Dωω /Dω , dk
(10.3.5)
have been used. Equation (10.3.3) is known as the nonlinear Schrödinger (NLS) equation. More explicitly, if the nonlinear dispersion relation is given by
ω = ω k, a2 , (10.3.6) and if we expand ω in a Taylor series about k = k0 and |a|2 = 0, we obtain 1 ∂ω |a|2 . (10.3.7) ω ≈ ω0 + (k − k0 )ω0 + (k − k0 )2 ω0 + 2 ∂|a|2 |a|2 =0 ∂ ∂ Replacing (ω − ω0 ) by i( ∂t ), k − k0 by −i( ∂x ), and assuming that the resulting operators act on the amplitude function a(x, t), it turns out that
1 i(at + ω0 ax ) + ω0 axx + γ|a|2 a = 0, 2
where γ=−
∂ω ∂|a|2
(10.3.8)
= const.
(10.3.9)
|a|2 =0
Equation (10.3.8) is known as the nonlinear Schrödinger equation. If we choose a frame of reference moving with the linear group velocity ω0 , that is, ξ = x − ω0 t and τ = t, the term involving ax will drop out from (10.3.8), and therefore, the amplitude a(x, t) satisfies the normalized nonlinear Schrödinger equation 1 iaτ + ω0 aξξ + γ|a|2 a = 0. 2
(10.3.10)
The corresponding dispersion relation is given by ω=
1 2 ω k − γa2 . 2 0
(10.3.11)
According to the stability criterion established by Whitham (1974), the wave modulation is stable if γω0 < 0 or unstable if γω0 > 0. To study the solitary wave solution, it is convenient to use the NLS equation in the standard form iψt + ψxx + γ|ψ|2 ψ = 0,
−∞ < x < ∞, t > 0.
(10.3.12)
10.3 The Nonlinear Schrödinger Equation and Solitary Waves
539
We then seek waves of permanent form by assuming the solution X = x − U t,
ψ = f (X)ei(mX−nt) ,
(10.3.13)
for some functions f and constant wave speed U to be determined, and where m, n are constants. Substitution of (10.3.13) in (10.3.12) gives
(10.3.14) f + i(2m − U )f + n − m2 f + γ|f |2 f = 0. We eliminate f by setting 2m − U = 0, and then write n = m2 − α, so that f can be assumed real. Thus, equation (10.3.14) becomes f − αf + γf 3 = 0.
(10.3.15)
Multiplying this equation by 2f and integrating, we find that f 2 = A + αf 2 −
γ 4 f ≡ F (f ), 2
(10.3.16)
where F (f ) ≡ (α1 − α2 f 2 )(β1 − β2 f 2 ), so that α = −(α1 β2 + α2 β1 ), A = α1 β1 , γ = −2(α2 β2 ), and the α’s and β’s are assumed to be real and distinct. Evidently, it follows from (10.3.16) that
f
X=
0
df (α1 − α2 f 2 )(β1 − β2 f 2 )
.
(10.3.17)
Setting (α2 /α1 )1/2 f = u in this integral, we deduce the following elliptic integral of the first kind (Dutta and Debnath 1965):
f
σX =
0
df (1 −
u2 )(1
− κ2 u2 )
,
(10.3.18)
where σ = (α2 β1 )1/2 and κ = (α1 β2 )/(β1 α2 ). Thus, the final solution can be expressed in terms of the Jacobian elliptic function u = sn(σX, κ). (10.3.19a) Thus, the solution for f (X) is given by f (X) =
α1 α2
1/2 sn(σX, κ).
(10.3.19b)
In particular, when A = 0, α > 0, and γ > 0, we obtain a solitary wave solution. In this case, (10.3.16) can be rewritten √ αX =
f
− 12 γ 2 f f 1− df. 2α 2
0
(10.3.20)
540
10 The Nonlinear Schrödinger Equation and Solitary Waves
Substitution of (γ/2α)1/2 f = sech θ in this integral gives the exact solution f (X) =
2α γ
1/2
√ sech α(x − U t) .
(10.3.21)
This represents a solitary wave that propagates without change of shape with constant velocity U . Unlike the solution of the KdV equation, the amplitude and the velocity of the wave are independent parameters. It is noted that the solitary wave exists only for the unstable case (γ > 0). This means that small modulations of the unstable wavetrain lead to a series of solitary waves. The well-known nonlinear dispersion relation for deep water waves is 1 (10.3.22) ω = gk 1 + a2 k 2 . 2 Therefore, ω0 =
ω0 , 2k0
ω0 =
ω0 , 4k02
1 and γ = − ω0 k02 , 2
(10.3.23)
and the NLS equation for deep water waves is obtained from (10.3.8) in the form ω0 1 ω0 i at + ax − 2 axx − ω0 k02 |a|2 a = 0. (10.3.24) 2k0 8k0 2 The normalized form of this equation in a frame of reference moving with the linear group velocity ω0 is 1 ω0 (10.3.25) iat − axx − ω0 k02 |a|2 a = 0. 8k02 2 Since γω0 = (ω02 /8) > 0, this equation confirms the instability of deep water waves. This is one of the most remarkable recent results in the theory of water waves. We next discuss the uniform solution and the solitary wave solution of the NLS equation (10.3.25). We look for solutions in the form
a(x, t) = A(X) exp iγ 2 t , X = x − ω0 t, (10.3.26) and substitute this in equation (10.3.25) to obtain the following equation: 8k 2 1 (10.3.27) AXX = 0 γ 2 A + ω0 k02 A3 . ω0 2 We multiply this equation by 2AX and, then, integrate to find
8 A2X = − A40 m2 + γ 2 k02 A2 + 2k04 A4 = A20 − A2 A2 − m2 A30 , (10.3.28) ω0 where A40 m2 is an integrating constant, 2k04 = 1, m 2 = 1 − m2 , and A20 = 4γ 2 /ω0 k02 (m2 − 2), which relates A0 , γ, and m.
10.4 Properties of the Solutions of the Nonlinear Schrödinger Equation
541
Finally, we rewrite (10.3.28) as A20 dX =
dA [(1 −
2 A2 )( A A20 A20
− m2 )]1/2
,
(10.3.29a)
or equivalently, A0 (X − X0 ) =
t
[(1 −
ds , − m2 )]1/2
s2 )(s2
s = (A/A0 ).
(10.3.29b)
This can readily be expressed in terms of the Jacobi dn function (Dutta and Debnath 1965): (10.3.30) A = A0 dn A0 (X − X0 ), m , where m is the modulus of the dn function. In the limit, m → 0, dnz → 1, and γ 2 → − 12 ω0 k02 A20 . Hence, the solution becomes 1 a(x, t) = A(t) = A0 exp − iω0 k02 A20 t . (10.3.31) 2 On the other hand, when m → 1, dnz → sech z, and γ 2 → − 14 ω0 k02 A20 . Therefore, the solitary wave solution is i (10.3.32) a(x, t) = A0 exp − ω0 k02 A20 t sech A0 (x − ω0 t − X0 ) . 4
10.4 Properties of the Solutions of the Nonlinear Schrödinger Equation We discuss several important properties of the nonlinear Schrödinger equation in the form (10.4.1) iuτ + βuξξ + γ|u|2 u = 0, where β and γ are real constants. When β = 12 ω0 , this equation reduces to (10.3.10). Simple solutions of equation (10.4.1) can always be obtained. However, the nature of the solutions depends on the signs of β and γ. If they are of the same sign, that is, if βγ > 0, the solution of (10.4.1), which tends to zero as |ξ| → ∞, represents a solitary wave solution
2ν −ν ξ exp(−iντ ), (10.4.2) u(ξ, t) = − sech γ β where ν is a constant parameter. It follows from the Galilean invariance of the Schrödinger equation that equation (10.4.1) is also invariant under the following transformations: iU 1 ˜ ξ = ξ − U τ, τ˜ = τ, u ˜ = u exp − ξ − Uτ , (10.4.3) 2β 2
542
10 The Nonlinear Schrödinger Equation and Solitary Waves
Fig. 10.1 An envelope soliton.
where U is another constant. Thus, equation (10.4.1) admits a solitary wave solution in the form 1 2
i U U 2ν 2 −ν ˜ − ξ exp i u(ξ, τ ) = − sech +2ν τ , (10.4.4) γ β 2β 2 2β where ν and U are independent constants. These solutions decrease very rapidly as |ξ| → ∞, so that they may be considered to represent solitary waves which are similar to KdV solitons. However, unlike the KdV equation, the nonlinear Schrödinger equation does not admit a soliton solution corresponding to a steady wave propagating with a constant velocity. In fact, the plane wave part represented by the exponential function and the amplitude of the sech profile propagate with different velocities. In this sense, the solution (10.4.4) is called the envelope soliton, as shown by Figure 10.1. It was first shown numerically by Yajima and Outi (1971) that, like the KdV soliton, the envelope soliton behaves like a particle. This was also proved analytically by Zakharov and Shabat (1972) by the inverse scattering method. An excellent experimental agreement with the sech-soliton solution (10.4.4) was found by Hammack as reported by Ablowitz and Segur (1979). Hammack’s experimental findings are shown in Figure 7.2 in Debnath (1994, p. 363). On the other hand, if u tends to a constant value u0 at infinity, then the plane wave solution exists in the form (10.4.5) u(ξ, τ ) = u0 exp i(ξ − δτ ) , where
δ = β2 − γu20 .
(10.4.6)
However, the plane wave solution is modulationally unstable. This can be shown by introducing two real functions ρ and σ through the expression
i √ σ dξ . (10.4.7) u = ρ exp 2β
10.4 Properties of the Solutions of the Nonlinear Schrödinger Equation
543
Substituting this expression in (10.4.1) gives ∂ ∂ρ + (ρσ) = 0 ∂τ ∂ξ and
(10.4.8)
∂σ ∂ρ 1 ∂ρ ∂σ ∂ 1 ∂ +σ = 2βγ + β2 . √ √ ∂τ ∂ξ ∂ξ ∂ξ ρ ∂ξ ρ ∂ξ
(10.4.9)
Since βγ > 0, in the long wavelength approximation, equations (10.4.8), (10.4.9) are equivalent to a hydrodynamic system with negative pressure. In other words, the system becomes elliptic. Thus, we consider that the perturbations modulate a constant amplitude ρ0 and phase σ0 so that (ρ, σ) = (ρ0 , σ0 ) + (δρ, δσ) exp i(κξ − Ωτ ) . (10.4.10ab) Substituting these perturbations in equations (10.4.8), (10.4.9) leads to an equation for the growth rate Ω in terms of the wavenumber κ as
1 (10.4.11) Ω = −σ0 κ ± (−2βγρ0 ) 2 κ + O κ3 . This shows that the perturbation grows exponentially as t → ∞ for small wavenumbers κ, provided that ρ0 is finite. On the other hand, if βγ > 0, that is, β and γ are of opposite signs, there exists a stable plane wave solution. In this case, the system of equations (10.4.8), (10.4.9) can be reduced to the KdV equation. This follows from the expansion of ρ and σ in terms of the small parameter ε about the constant state, so that we can write ρ = ρ0 + ερ1 + ε2 ρ2 + · · ·
and σ = σ0 + εσ1 + ε2 σ2 + · · · .
These results combined with the transformation √ 1 ξ˜ = ε ξ − σ0 + (−2βγρ0 ) 2 τ ,
τ˜ = ετ
(10.4.12ab)
(10.4.13)
reduce the system (10.4.8), (10.4.9) to the KdV equation for ρ1 and σ1 . Thus, when βγ > 0, the above solution is often called a bright soliton (or envelope soliton). The bright pulse arises when |u|2 increases from a finite value at infinity and, subsequently, returns to the same value. When βγ < 0, the solution is of the expansive type in which |u|2 decreases from a finite value at infinity and subsequently returns to the same state. This is often called a dark soliton (or dark pulse). So far, the self-modulation of a single mode governed by the dispersion relation has been described. However, when different modes with amplitudes of the same order O(ε) coexist and undergo mutual interactions and self-modulation, it is possible to split the system of equations into independent nonlinear Schrödinger equations similar to the case of long waves (Oikawa and Yajima 1974a, 1974b). Consequently, envelope solitons associated with various modes, moving separately at the beginning, preserve their identities and, then, propagate as envelope solitons, even after the mutual interactions.
544
10 The Nonlinear Schrödinger Equation and Solitary Waves
Zakharov and Shabat (1972) discovered that the inverse scattering method gives an exact solution of the initial-value problem for the nonlinear Schrödinger equation. Based on the initial data, which tend to zero rapidly at infinity, their remarkable analysis can be summarized as follows. (i) An initial wave envelope pulse of arbitrary shape eventually disintegrates into a number of solitons of shorter scales and an oscillatory tail. Each soliton is a permanent, progressive wave. The number and structure of these solitons and the nature of the tail are completely described by the initial data. (ii) These soliton solutions are definitely stable in the sense that they undergo nonlinear interaction and, then, emerge from the interaction without permanent changes except for a possible change in position and phase. (iii) The oscillatory tail is relatively small and disperses linearly with amplitude 1 which decays like t− 2 , as t → ∞. Another significant property of the solutions of the nonlinear Schrödinger equation is the recurrence phenomenon. In a conservative system, the existence of invariants plays a fundamental role in establishing recurrence phenomena, as shown by Gibbons et al. (1977) and by Thyagaraja (1979, 1981, 1983). As demonstrated by these authors, the recurrence phenomenon is typical only for bounded or periodic domains. Thyagaraja considered the initial-value problem for the nonlinear Schrödinger equation iψt = ψxx + γ|ψ|2 ψ (10.4.14) in the periodic domain 0 ≤ x ≤ 1 with periodic boundary conditions and a real constant γ which can be positive or negative. In many physical problems, γ is positive. We assume the periodic boundary conditions for t > 0,
ψ(0, t) = 0 = ψ(1, t)
(10.4.15)
or for t > 0,
(10.4.16)
for all x ∈ [0, 1].
(10.4.17)
ψx (0, t) = 0 = ψx (1, t) and the initial condition ψ(x, 0) = Ψ (x)
Without proof, we assume that this initial-boundary problem for equation (10.4.14) possesses a smooth solution which is uniquely determined by the initial data Ψ (x). To investigate the qualitative properties of the solution, Thyagaraja derived certain a priori bounds involving the integral invariants associated with equation (10.4.14). Two invariants of the problem which are constants of the motion are given by 1 ψ(x, t)2 dx, I(t) = (10.4.18) 0
and
1
J(t) = 0
ψx (x, t)2 dx − γ 2
0
1
ψ(x, t)4 dx.
(10.4.19)
10.4 Properties of the Solutions of the Nonlinear Schrödinger Equation
545
However, Zakharov and Shabat (1972) proved that this problem has an infinite set of integral invariants, provided ψ(x, t) is sufficiently smooth. We consider any function ψ(x, t), not necessarily a solution of (10.4.14), which is defined for |t| > 0 and sufficiently smooth. We assume that ψ(x, t) satisfies the given boundary conditions and evolves in time, so that functions I(t) and J(t), defined by (10.4.18), (10.4.19), are constants in time. We then find bounds of the Rayleigh quotient Q(t) defined by
1
ψx (x, t)2 dx.
Q(t)I(t) =
(10.4.20)
0
Denoting I(0) = I0 and J(0) = J0 , we find that Q(t) <
J0 , I0
(10.4.21)
provided that γ < 0. For γ > 0, we assume that |ψ(x, t)|2 is minimal at x = x0 for any t, and hence, we obtain x ψ(x, t)2 = ψ(x0 , t)2 + 2 ψψx dx, (10.4.22) x0
which leads to the inequality ψ(x, t)2 ≤ ψ(x0 , t)2 + 2
1
ψψx dx.
(10.4.23)
0
Obviously, ψ(x0 , t)2 ≤
1
ψ(x0 , t)2 dx ≤
0
1
ψ(x, t)2 dx,
0
and the Schwarz inequality
ψψ0 dx ≤ I(t) Q(t).
1
0
Multiplying (10.4.23) by |ψ(x, t)|2 and integrating gives
ψ(x, t)4 dx ≤ I 2 + 2I 2 Q(t).
1
(10.4.24)
0
Substituting in (10.4.19) yields the quadratic inequality J0 γ Q(t) ≤ I0 + γI0 Q(t). + I0 2 Finally, (10.4.23) can be reorganized to obtain ψ(x, t)2 ≤ I0 + 2I0 Q(t).
(10.4.25)
(10.4.26)
546
10 The Nonlinear Schrödinger Equation and Solitary Waves
Since, Q(t) ≥ 0 by definition, it is easy to solve the inequality (10.4.25) and obtain the following results: Q(t) ≤ M 2 (I0 , J0 , γ), 2
ψ 2∞ ≡ max ψ(x, t) ≤ I0 (1 + 2M ), 0≤x≤1
where M is the positive root of the quadratic equation J0 γ M 2 − γI0 M − + I0 = 0. I0 2
(10.4.27) (10.4.28)
(10.4.29)
To examine the major implications of the a priori bounds in (10.4.27), (10.4.28), we introduce the concept of Lagrangian stability. A solution ψ(x, t) of (10.4.14) is said to be Lagrangian stable if there exists a constant K independent of t, but possibly dependent on the initial data such that ψ(x, t)2 = max ψ(x, t)2 ≤ K, ∞ 0≤x≤1
(10.4.30)
for all |t| ≥ 0. Clearly, it follows from (10.4.28) with K = I0 (1+2M ) that any solution ψ(x, t) is Lagrangian stable. To give an interpretation of the inequality (10.4.27), we expand ψ(x, t) in a Fourier series in x, so that ψ(x, t) =
∞
an (t) exp(2nπix),
(10.4.31)
n=−∞
where I0 =
∞ an (t)2 n=−∞
and Q(t) =
4π 2 I0
∞
2 n2 an (t) .
(10.4.32)
n=−∞
In analogy with quantum physics, we may interpret Q(t)/4π 2 as the instantaneous average of n2 , and (M/2π) can be interpreted as an upper bound to the rms√value 1 ) Q is of n, the number of modes carrying the “wave energy” I0 . The quantity ( 2π called the number of effective modes and is denoted by Neff . The inequality (10.4.27) M ), which depends only on the initial values of I0 and says that Neff is bounded by ( 2π J0 of the integral invariants and on the interaction strength γ. Thus, we have proved that every motion described by equation (10.4.14) with given I0 and J0 can be assigned Neff as a measure of the number of effective modes. It is then straightforward to apply Birkhoff’s theorem (1927) to conclude that, in general, dynamical systems with a finite number of degrees of freedom and bounded motions are “generically” recurrent. It is interesting to compare the results of this analysis with numerical simulations and experiments. Yuen and Ferguson (1978a, 1978b) investigated spatially periodic
10.5 Conservation Laws for the NLS Equation
547
solutions of (10.4.14). They numerically evolved initial data for which the linearized form of (10.4.14) is unstable to the modulational instability of Benjamin and Feir (1967). The longtime behavior of these solutions reveals that the energy is shared effectively between a finite number of wave modes. The number depends on the initial data. Moreover, the solution reconstructed itself after appearing to show a tendency to break up due to linear instability (Yuen and Lake 1980). Both the failure of the energy to thermalize and the tendency to recur have been known since the classic work of Fermi et al. (1955). They observed a remarkable recurrence phenomenon, known as the Fermi–Pasta–Ulam (FPU) recurrence. Initially, if energy is given to the lowest mode with the lowest frequency, then several higher modes are excited by the nonlinear effects, but after a suitable lapse of time during which a flowing back and forth among several low-order modes takes place, the energy eventually returns to the mode comprising the initial state. In other words, the system does not approach thermal equilibrium by the principle of the equipartition of energy due to the nonlinear effects, but the FPU recurrence is observed, and consequently, the nonergodicity of the system can be demonstrated. Subsequently, Zabusky and Kruskal (1965) and Fornberg and Whitham (1978) reported similar observations. From a series of computational and experimental works by Yuen and Lake (1975) and Lake et al. (1977), it has been confirmed that recurrence is a generic phenomenon for any physical problem described by the nonlinear Schrödinger equation. At the same time, the above mathematical analysis is now accepted as conclusive evidence of the FPU recurrence phenomenon. For a long time it was mistakenly thought that the recurrence property of the NLS equation is attributable to the fact that it admits a multisoliton solution and can be solved exactly by the inverse scattering method. However, the above discussion shows, beyond any doubt, that the recurrence of solutions has nothing to do with soliton solutions or exact integrability.
10.5 Conservation Laws for the NLS Equation Zakharov and Shabat (1972) proved that equation (10.3.12) has an infinite number of polynomial conservation laws. These have the form of an integral, with respect to x, of a polynomial expression in terms of the function ψ(x, t) and its derivatives with respect to x. These laws are somewhat similar to those already proved for the KdV equation. Therefore, the proofs of the conservation laws are based on similar assumptions used in the context of the KdV equation (see Section 9.6). We prove here three conservation laws for the NLS equation (10.3.12): ∞ |ψ|2 dx = const. = C1 , (10.5.1) −∞ ∞
i ψψ x − ψψx dx = const. = C2 , (10.5.2) −∞ ∞
1 |ψx |2 − γ|ψ|4 dx = const. = C3 . (10.5.3) 2 −∞
548
10 The Nonlinear Schrödinger Equation and Solitary Waves
We multiply (10.3.12) by ψ and its complex conjugate by ψ and subtract the latter from the former to obtain i
d d ψψ + ψx ψ − ψ x ψ = 0, dt dx
(10.5.4)
where the bar denotes the complex conjugate. Integration with respect to x in −∞ < x < ∞ gives d ∞ i |ψ|2 dx = 0. dt −∞ This proves result (10.5.1). We multiply (10.3.12) by ψ x and its complex conjugate by ψx and, then, add the resulting expressions to obtain
(10.5.5) i ψt ψ x − ψ t ψx + ψxx ψ x + ψ xx ψx + γ|ψ|2 ψψ x + ψψx = 0. We differentiate (10.3.12) and its complex conjugate with respect to x, multiply the former by ψ and the latter by ψ, and then, add them together. This leads to the result
i ψψxt − ψψ xt + ψxxx ψ + ψ xxx ψ
(10.5.6) + γ ψ |ψ|2 ψ x + ψ |ψ|2 ψ x = 0. If we subtract (10.5.6) from (10.5.5) and then simplify, it turns out that i
d d d ψψx − ψψ x = ψψxx + ψψ xx ψx ψ x + dt dx dx 2 d d ψx ψ x + γ ψψ − dx dx d d = ψψxx + ψψ xx + γ |ψ|4 . dx dx
Integrating this result with respect to x yields d ∞ i ψψx − ψψ x dx = 0. dt −∞ This proves result (10.5.2). We multiply (10.3.12) by ψ t and its complex conjugate by ψt and add the resulting equations to derive
2 ψ t ψxx + ψ xx ψt + γ ψ 2 ψψt + ψ ψψt = 0,
or equivalently, γ d 2 2 d d ψx ψ x + ψ ψ = 0. ψ t ψx + ψt ψ x − dx dx 2 dt
10.5 Conservation Laws for the NLS Equation
Integrating this result with respect to x gives
∞ γ 4 d d ∞ 2 ψ t ψx + ψt ψ x dx = 0. |ψx | − |ψ| dx = dt −∞ 2 −∞ dx
549
(10.5.7)
This gives (10.5.3). The preceding three conservation integrals have a simple physical meaning. In fact, the constants of motion C1 , C2 , and C3 are related to the number of particles, the momentum, and the energy of a system governed by the nonlinear Schrödinger equation. The analysis of this section reveals several remarkable features of the nonlinear Schrödinger equation. This equation can also be used to investigate instability phenomena in many other physical systems. Like the various forms of the KdV equation, the NLS equation arises in many physical problems, including nonlinear water waves and ocean waves, waves in plasma, propagation of heat pulses in a solid, selftrapping phenomena in nonlinear optics, nonlinear waves in a fluid-filled viscoelastic tube, and various nonlinear instability phenomena in fluids and plasmas. Whitham’s equations for the slow modulation of the wave amplitude a and the wavenumber k in the case of two-dimensional deep water waves are given by 2 ∂ a ∂ a2 + = 0, (10.5.8) C ∂t ω0 ∂x ω0 ∂k ∂ω + = 0, (10.5.9) ∂t ∂x √ where ω0 = gk is the first-order approximation for the wave frequency ω and C = (g/2ω0 ) is the group velocity. Chu and Mei (1970, 1971) observed that certain terms of the dispersive type, neglected in Whitham’s equations to the same order of approximation, must be included to extend the validity of these equations. Whitham’s theory is based on the direct use of Stokes’ dispersion relation for a uniform wavetrain, 1 2 2 2 ω = ω0 1 + ε a k , (10.5.10) 2 whereas Chu and Mei added terms of higher derivatives and dispersive type to the expression for ω, so that
a 1 2 2 ÷ 2ω0 a . (10.5.11) ω = ω0 1 + ε2 a k + 2 ω0 tt They used the expression (10.5.11) to transform (10.5.8), (10.5.9) in a frame of reference moving with the group velocity C and obtained the following nondimensional equations: ∂ 2 ∂a2 + a φx = 0, ∂t ∂x
(10.5.12)
550
10 The Nonlinear Schrödinger Equation and Solitary Waves
−2
∂2φ a2 ∂ axx + −φ2x + + = 0, ∂x∂t ∂x 4 16a
(10.5.13)
where we have used Chu and Mei’s result W = −2φx and φ is a small phase variation. Integrating (10.5.13) with respect to x and setting the constant of integration to be zero gives 1 1 axx φt + φ2x − a2 − = 0. (10.5.14) 2 8 32a A transformation Ψ = a exp(4iφ) is used to simplify (10.5.12) and (10.5.14), which reduces to the nonlinear Schrödinger equation 1 1 iΨt + Ψxx + Ψ |Ψ |2 = 0. 8 2
(10.5.15)
This equation has also been derived and exploited by several authors, including Benney and Roskes (1969), Hasimoto and Ono (1972), and Davey and Stewartson (1974) to examine the nonlinear evolution of Stokes’ waves on water. An alternative way to study nonlinear evolution of two- and three-dimensional wavepackets is to use the method of multiple scales in which the small parameter ε is explicitly built into the expansion procedure. The small parameter ε characterizes the wave steepness. This method has been employed by several authors in various fields and has also been used by Hasimoto and Ono (1972) and Davey and Stewartson (1974).
10.6 The Inverse Scattering Method for the Nonlinear Schrödinger Equation Zakharov and Shabat (1972) developed an ingenious method for solving the nonlinear Schrödinger equation in the form iut + uxx + 2|u|2 u = 0
(10.6.1)
by considering the spectral problem v1x = −iζv1 + qv2 , v2x = iζv2 + rv1 .
(10.6.2) (10.6.3)
In matrix notation, they take the form ∂ −q −iζ q ∂x v = −iζv, (10.6.4ab) vx = v, or ∂ r iζ r − ∂x
where v is the vector vv12 , bounded functions q(x) and r(x) (not necessarily real) are potentials, and ζ is the eigenvalue.
10.6 The Inverse Scattering Method for the Nonlinear Schrödinger Equation
551
The most general linear time dependence is given by v1t = Av1 + Bv2 , v2t = Cv1 + Dv2 ,
(10.6.5) (10.6.6)
where A, B, C, and D are scalar functions independent of v1 and v2 to be determined. To establish the relation between the pair of equations (10.6.2) and (10.6.3) and the Schrödinger scattering problem, we first differentiate (10.6.3) with respect to x to obtain v2xx = iζv2x + rx v1 + rv1x , (10.6.7) provided that rx exists. By equations (10.6.2) and (10.6.3), equation (10.6.7) becomes v2xx = iζv2x + rx v1 + r(−iζv1 + qv2 ) 1 = iζv2x + (rx − iζr)(v2x − iζv2 ) + rqv2 , r or equivalently, v2xx −
rx rx v2x − qr − iζ − ζ 2 v2 = 0. r r
(10.6.8)
The special choice r = −1 gives the Schrödinger scattering problem for v2
v2xx + ζ 2 + q v2 = 0. (10.6.9) When ζ 2 = λ and q = −u, the system (10.6.4ab) reduces to the time-independent Schrödinger equation required for solving the KdV equation which was discussed in Chapter 9. To determine the coefficients involved in (10.6.5), (10.6.6), we use cross-differentiation of (10.6.2), (10.6.3) and (10.6.5), (10.6.6) combined with the assumption that the eigenvalues are invariant in time (ζt = 0), that is, (vjx )t = (vjt )x for j = 1, 2, leading to the following equations for A, B, C, and D: Ax = qC − rB, Bx + 2iζB = qt − (A − D)q, Cx − 2iζC = rt + (A − D)r, −(D)x = qC − rB. Without loss of generality, we can take −D = A so that the preceding system of equations becomes Ax = qC − rB, Bx + 2iζB = qt − 2Aq, Cx − 2iζC = rt + 2Ar.
(10.6.10) (10.6.11) (10.6.12)
552
10 The Nonlinear Schrödinger Equation and Solitary Waves
Since the Schrödinger equation is second order in x, we may expect that the series expansions in ζ for A, B, and C include only terms through ζ 2 , and so, we write an exact truncated power series A=
2 n=0
An ζ n ,
B=
2
Bn ζ n ,
C=
n=0
2
Cn ζ n .
(10.6.13)
n=0
Substituting these expressions in (10.6.10)–(10.6.12) and equating coefficients of powers of ζ gives An , Bn , and Cn where n = 0, 1, 2. The coefficients of ζ 3 in (10.6.11) and (10.6.12) give B2 = C2 = 0. For coefficients of ζ 2 , equation (10.6.10) yields A2 = a2 = const. Equations (10.6.11) and (10.6.12) give (B1 , C1 ) = ia2 (q, r). For the coefficients ζ, equation (10.6.10) gives A1 = a1 = const. For simplicity, we set a1 = 0. However, if a1 = 0, a more general evolution equation can be obtained. Then, (10.6.11), (10.6.12) give (B0 , C0 ) = 12 a2 (−qx , rx ). Finally, for the coefficients ζ 0 , (10.6.10) gives A0 = 12 a2 qr + a0 , where a0 is a constant which is set equal to zero. Consequently, equations (10.6.11) and (10.6.12) for coefficients of ζ 0 give 1 − a2 qxx = qt − a2 q 2 r, 2 1 a2 rxx = rt + a2 qr 2 . 2
(10.6.14) (10.6.15)
This represents a coupled pair of nonlinear evolution equations. In particular, when r = ±q ∗ , equations (10.6.14) and (10.6.15) are compatible provided that a2 = ia, where a is real. If we take a = 2, we obtain the equation iqt = qxx ± 2q 2 q ∗ .
(10.6.16)
This equation with the positive sign is called the nonlinear Schrödinger equation which admits a special soliton solution, while the equation with the negative sign does not give any soliton solution for potentials decaying rapidly as x → ∞ since the spectral operator in (10.6.2) and (10.6.3) is self-adjoint (or Hermitian).
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics Example 10.7.1 (The Nonlinear Schrödinger Equation for Deep Water and the Benjamin–Feir Instability). One of the simplest solutions of the nonlinear Schrödinger equation (10.3.25) is given by (10.3.31), that is, 1 A(t) = A0 exp − iω0 k02 A20 t , (10.7.1) 2 where A0 is a constant. This essentially represents the fundamental component of the Stokes wave. We consider a perturbation of (10.7.1) and express it in the form
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics
a(x, t) = A(t) 1 + B(x, t) ,
553
(10.7.2)
where B(x, t) is the perturbation function. Substituting this result in (10.3.25) gives ω0 ABxx i(1 + B)At + iABt − 8k02 1 = ω0 k02 A20 (1 + B) + BB ∗ (1 + B) 2 + (B + B ∗ )B + (B + B ∗ ) A, (10.7.3) where B ∗ (x, t) is the complex conjugate of the perturbed function B(x, t). Neglecting the squares of B and simplifying, equation (10.7.3) reduces to 1 ω0 Bxx = ω0 k02 A20 (B + B ∗ ). iBt − (10.7.4) 8k02 2 We look for a solution for the perturbed quantity B(x, t) in the form B(x, t) = B1 exp(Ωt + ix) + B2 exp(Ω ∗ t − ix),
(10.7.5)
where B1 and B2 are complex constants, is a real wavenumber, and Ω is a growth rate (possibly a complex quantity) to be determined. Substituting the solution for B in (10.7.4) yields a pair of coupled equations: ω0 2 1 B1 − ω0 k02 A20 (B1 + B2∗ ) = 0, iΩ + (10.7.6) 8k02 2 ω0 2 1 B2 − ω0 k02 A20 (B1∗ + B2 ) = 0. (10.7.7) iΩ ∗ + 8k02 2 We take the complex conjugate of (10.7.7) to transform it into the form ω0 2 1 −iΩ + B2∗ − ω0 k02 A20 (B1 + B2∗ ) = 0. 8k02 2
(10.7.8)
The pair of linear homogeneous equations (10.7.6) and (10.7.8) for B1 and B2∗ admits a nontrivial eigenvalue for Ω provided iΩ + ω0 2 − 1 ω k 2 A2 1 2 2 − ω k A 2 0 0 0 0 0 0 2 2 8k0 = 0, (10.7.9) ω0 2 1 1 2 2 2 2 − 2 ω0 k 0 A 0 iΩ + 8k2 − 2 ω0 k0 A0 0
which is equivalent to Ω2 =
2 1 ω0 2 k02 A20 − 2 . 2 2k0 8k0
(10.7.10)
The growth rate Ω is purely imaginary or real (and positive) depending on whether 2 > 8k04 A20 or 2 < 8k04 A20 . The former case represents a wave solution for B,
554
10 The Nonlinear Schrödinger Equation and Solitary Waves
and the latter corresponds to the Benjamin–Feir (or modulational) instability with a criterion in terms of the nondimensional wavenumber ˜ = (/k0 ) as ˜2 < 8k02 A20 .
(10.7.11)
Thus, the range of instability is given by √ 0 < ˜ < ˜c = 2 2k0 A0 .
(10.7.12)
˜ the maximum instability occurs at ˜ = ˜max = 2k0 A0 , Since Ω is a function of , with a maximum growth rate given by (Re Ω)max =
1 ω0 k02 A20 . 2
(10.7.13)
To establish the connection with the Benjamin–Feir instability, we have to find the velocity potential for the fundamental wave mode multiplied by exp(kz). It turns out that the term proportional to B1 is the upper sideband, whereas that proportional to B2 is the lower sideband. The main conclusion of the preceding analysis is that Stokes water waves are definitely unstable. Example 10.7.2 (The Nonlinear Klein–Gordon Equation). We consider the nonlinear Klein–Gordon equation in the form ψtt − (ψxx + ψyy ) + m2 ψ + γψ 3 = 0,
x ∈ R, 0 ≤ y ≤ a,
(10.7.14)
where m and γ are constants. We assume the boundary conditions ψ = 0 for y = 0 and y = a.
(10.7.15)
For the linear case, we assume ψ(x, y, t) = Ψ (y) exp i(kx − ωt)
(10.7.16)
to obtain α2 = ω 2 − k2 − m2 .
Ψ (y) = A sin(αy) + B cos(αy),
(10.7.17)
Since B = 0 by (10.7.15), from the boundary conditions, we obtain α =ω −k −m = 2
2
2
2
nπ a
2 ,
n = 1, 2, 3, . . . ,
(10.7.18)
so that the group velocity C = ω (k) = ωk and ω = ω(k, n). Hence, there exist various modes dependent on n. We consider a wave with n = 1, and its amplitude varies slowly due to nonlinearity, so that slow variables ξ and τ defined by (10.2.5ab) can be used. We next assume the following form of the solution:
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics ∞
ψ= ψ
(m)
=
555
εm ψ (m) ,
m=1 ∞
(m) ψ (ξ, τ, y) exp i(kx
(10.7.19ab)
− ωt) .
=−∞
Retaining only terms of the first order in ε, we obtain (1)
ψ1 = φ(ξ, τ ) sin(αy),
(1)
ψ
= 0,
|| = 1,
(10.7.20)
and the second-order terms in ε give (1)
2i(k − ωC)
∂ψ ∂ξ
(2) ∂ 2 ψ + 2 ω 2 − 2 k 2 − m2 ψ = 0, ∂2y (2)
+
where the first term vanishes because C = the terms of the third order in ε yield (3)
∂ 2 ψ1 ∂y 2
k ω
(2)
so that ψ
(10.7.21)
= 0 for || = 1. For = 1,
∂2φ ∂φ (3) + α2 ψ1 + 2iω sin(αy) + 1 − C2 ∂τ ∂ξ 2
− 3γ|φ|2 φ sin3 (αy) = 0.
(10.7.22)
Multiplying this equation by sin(αy) and integrating with respect to y from 0 to a causes the first and second terms to vanish. Consequently, equation (10.7.22) gives 2 ω − k2 9γ 2 (10.7.23) φξξ − |φ| φ = 0. iφτ + 2ω 3 8ω This is a nonlinear Schrödinger equation for the amplitude φ(ξ, τ ). Since ω2 − k 2 > 0, a modulational instability occurs when γ > 0. Denoting the initial amplitude by A0 , the solitary wave solution is given by
9γA20 (1) ψ1 (ξ, τ, y) = A0 sin(αy) exp (iτ ) 16ω 12
9γ × sech Aω − ξ . (10.7.24) 8(ω 2 − k 2 ) Example 10.7.3 (Motion of a Vortex Filament and the Nonlinear Schrödinger Equation). The motion of a very thin isolated vortex filament X = X(x, t) of radius ε in an incompressible unbounded fluid by its own induction was described asymptotically by Hasimoto (1972) in the form ∂X = Gκb, ∂t
(10.7.25)
where s is the length measured along the filament, t is the time, κ is the curvature, b is the unit vector along the binormal, and G is the coefficient of local induction
556
10 The Nonlinear Schrödinger Equation and Solitary Waves
G=
Γ −1 log ε + O(1) , 4π
(10.7.26)
which is proportional to the circulation Γ of the filament and may be treated as constant if the slow variation of the logarithm is compared with that of its argument. The local motion is approximated by that of a thin circular ring with the same curvature. Under this approximation, the tangential motion along the filament due to stretching is neglected, even though it is very important in many cases. Introducing a suitable choice of the units of length and time transforms (10.7.25) into the nondimensional form ˙ = κb, X (10.7.27) where the dot denotes a partial derivative with respect to time. We next include the equations of differential geometry, that is, the Serret–Frenet formulas for the tangent vector t, the principle normal vector n, and the binormal b (t = n × b and t, n, and b form a right-handed system), t = κn,
n = τ b − κt,
and b = −τ n,
(10.7.28)
where a prime denotes a partial derivative with respect to arc length s, κ is the curvature, and τ is the torsion of the curve. When κ and τ are specified at each point of the curve, the shape of the curve, except for its location, is uniquely determined. We prove that the complex function
s τ (s , t) ds , ψ(s, t) = κ(s, t) exp i 0
which contains both κ and τ , and hence, completely determines the shape of the filament and satisfies the nonlinear Schrödinger equation. We first combine the last two of the Serret–Frenet formulas (10.7.28) into the complex form (n + ib)s + iτ (n + ib) = −κt.
(10.7.29)
Introducing N = (n + ib) exp i
s
τ ds ,
ψ = κ exp i
0
s
τ ds ,
(10.7.30ab)
0
we obtain
N = −ψt,
(10.7.31)
and the first equation of (10.7.28) becomes t =
1 ∗ ψ N + ψN∗ , 2
(10.7.32)
where the asterisk denotes the complex conjugate. This displacement of the position vector along the filament gives a unit tangent vector to the curve, that is,
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics
557
(10.7.33)
Xs = t. To find the time derivatives of t and N, we proceed as follows: ∂t ∂ = (Xs ) = (κb)s = κs b − κτ n. ∂t ∂t
(10.7.34)
Since κs b − κτ n = Re κs (b + in) + iκτ (b + in) = Re i(κs + iκτ )(n − ib) i = Re iψs N∗ = ψs N∗ − ψs∗ N , 2 1 ∂ ∗ ∗ t = i ψ s N − ψs N . ∂t 2
(10.7.35) (10.7.36)
In general, the time derivative of N is a linear combination of N, N∗ , and t, and hence, we can write ∂N = αN + βN∗ + γt, (10.7.37) ∂t where α, β, and γ are constants to be determined. It follows from the definition of N and the fact that t, n, and b are orthogonal that N · N = t · N = 0 and
N · N∗ = 2.
(10.7.38)
We take the dot product of (10.7.37) with each of N, N∗ , and t to derive 2β = N ·
1 ∂N = (N · N) = 0, ∂t 2
2α = N∗ ·
∂N . ∂t
(10.7.39)
Consequently,
∂ N · N∗ = 0, (10.7.40) ∂t and hence, α must be purely imaginary, so that α = iR, where R is real. Finally, ∂ (N · t) = 0, since ∂t 2(α + α∗ ) =
γ =t·
∂N ∂t i = −N · = − N · ψs N∗ − ψs∗ N , ∂t ∂t 2
(10.7.41)
so that γ = −iγs . It turns out that ∂N = i(RN − ψs t). ∂t
(10.7.42)
The time derivative of (10.7.31) and the s derivatives of (10.7.42) give i ∗ ∂ψ ψψ N − ψψs N∗ − t , 2 s ∂t i i = N iRs − ψs ψ ∗ − ψψs∗ N − i(Rψ + ψss )t. 2 2
Nst =
(10.7.43)
Nts
(10.7.44)
558
10 The Nonlinear Schrödinger Equation and Solitary Waves
Equating the components of these two expressions yields 1 ∂R = (ψ ∗ ψs + ψψs∗ ) ∂s 2 and
∂ψ ∂2ψ − i 2 − iRψ = 0. ∂t ∂s Integrating (10.7.45) gives the value of R as R=
1 ∗ ψψ + A(t) , 2
(10.7.45)
(10.7.46)
(10.7.47)
where A(t) arises due to integration. Substituting the value of R in (10.7.46) leads to the nonlinear Schrödinger equation iψt + ψss +
1 2 |ψ| + A(t) ψ = 0, 2
where A(t) can be eliminated by the transformation
1 i t A(t ) dt . u = ψ exp − 2 2 0
(10.7.48)
(10.7.49)
This transformation is nothing but a shift of the origin of integration in (10.7.30ab). Therefore, without loss of generality, we may set A = 0 in (10.7.48). Consequently, (10.7.48) reduces to the standard NLS equation 1 iψt + ψss + |ψ|2 ψ = 0. 2
(10.7.50)
We obtain a solitary wave solution which propagates steadily with a constant velocity c along the filament which is a straight line at infinity, that is, κ→0
as s → ∞.
(10.7.51)
Thus, in the frame of reference, κ and τ are functions of ξ = s − ct, that is,
s (10.7.52) ψ = κ(ξ) exp i τ (ξ) ds . 0
The real and imaginary parts of (10.7.48) give, respectively, 1 −cκ τ (ξ) − τ (−ct) = κ − κτ 2 + κ κ2 + A 2 and
cκ = 2κ τ + κτ .
(10.7.53)
(10.7.54)
Integrating (10.7.54) and using (10.7.51) to determine the constant of integration gives (c − 2τ )κ = 0, which yields τ = τ0 = 12 c since κ = 0. This means that the
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics
559
torsion is constant along the vortex filament and the velocity of propagation along the filament is twice the torsion. The shape of the vortex filament corresponding to the single solitary wave of the NLS equation, τ = 1, κ = 2 sech s, is shown in Figure 10.2. Several other figures have been presented by Hasimoto (1972). Finally, we use (10.7.51) to integrate (10.7.53) so that the solution becomes κ(ξ, t) = 2ν sech(νξ) = 2ν sech ν(s − ct) , (10.7.55) provided that A is a constant related to τ0 and ν by A = 2(τ02 − ν 2 ). The actual shape of the filament is obtained from (10.7.55) by setting τ0 = 12 c. Physically, the solution represents a helical curve of constant torsion τ0 with a curvature κ that decreases from a maximum value 2ν at the point s = ct to zero, as |s| → ∞. This single loop of helical motion moves along the vertex line with a velocity c = 2τ0 . Example 10.7.4 (Langmuir Solitons and Their Instability). We consider one-dimensional Langmuir waves propagating in a plasma of cold ions and hot electrons. We assume that the electrons respond adiabatically to high frequency motions. Using the standard notation for various quantities, which are similar to those involved in Example 9.3.1, the equations of motion for the electrons are given by ∂ ∂ne + (ne ue ) = 0, ∂t ∂x 2 3vT e ∂ne ∂ue eE ∂ue − + ue =− , ∂t ∂x me ne ∂x ∂E = 4πe(ni − ne ). ∂x
(10.7.56) (10.7.57) (10.7.58)
We decompose the various quantities into a high-frequency part and a low−1 −1 frequency part characterized by the time scales ωpe and ωpi , respectively, so that ne = n0 + n + neh ,
ni = n0 + n ,
E = Eh + E ,
(10.7.59)
where neh and Eh refer to the high-frequency perturbations and n and E denote the low-frequency perturbations. Assuming charge neutrality in the low-frequency motions, equations (10.7.56)–(10.7.58) reduce to the form ∂ ∂neh + (n0 + n )ue = 0, ∂t ∂x ∂ue eEh 3vT2 e ∂neh =− , − ∂t me (n0 + n ) ∂x ∂Eh = −4πeneh . ∂x Eliminating ue from (10.7.60) and (10.7.61) gives
e ∂Eh 3vT2 e ∂ 2 neh ∂ 2 neh + = 0, − (n0 + n ) ∂t2 me ∂x (n0 + n ) ∂x2
(10.7.60) (10.7.61) (10.7.62)
(10.7.63)
560
10 The Nonlinear Schrödinger Equation and Solitary Waves
Fig. 10.2 (a) Shape of the vortex filament, (b) projection of the curve on the (x, z)-plane, and (c) projection of the curve on the (x, y)-plane.
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics
561
which is, by (10.7.62), 2 n ∂ 2 Eh 2 2 ∂ Eh + ω − 3v = 0, 1 + E h pe Te ∂t2 n0 ∂x2 where 2 ωpe =
(10.7.64)
4πe2 n0 . me
Equation (10.7.64) describes the trapping of a Langmuir wavepacket in density cavities. We assume that Eh can be decomposed into a fast time-varying component and a slow modulation, that is, we seek a solution in the form Eh (x, t) = E(x, t) exp(−iωpe t),
(10.7.65)
where E(x, t) has a slow modulation in time t. Substituting (10.7.65) in (10.7.64) gives an equation for the modulation function E(x, t), ∂E ∂2E 3 1 2 n E. (10.7.66) ωpe + vT2 e 2 = iωpe ∂t 2 ∂x 2 n0 This equation is based on the hypothesis that the effects of nonlinearity and finite temperature appear as corrections of the same order to the linear dispersion relation of cold plasmas near cutoff values of the wavenumber. Based on the assumption that electrons respond isothermally to low-frequency motions, the low-frequency motions of plasma are given by vT2 e e ∂ue ∂n E = − − ve , me (n0 + n ) ∂x ∂x ∂ ∂n + (n0 + n )ui = 0, ∂t ∂x ∂ui e − E = 0, ∂t mi where electron inertia is ignored. Eliminating ui from (10.7.68), (10.7.69) yields
∂ e ∂ 2 n + + n ) E (n 0 = 0. ∂t2 ∂x me
(10.7.67) (10.7.68) (10.7.69)
(10.7.70)
We next use (10.7.67) to eliminate E from (10.7.70), so that equation (10.7.70) becomes 2 n0 me ∂ ∂ue ∂ 2 n 2 ∂ n − c = u . (10.7.71) e s ∂t2 ∂x2 mi ∂x ∂x −1 e , the low-frequency contribution of (ue ∂u Averaging over the time scale ωpe ∂x ) is given by
562
10 The Nonlinear Schrödinger Equation and Solitary Waves
2 ∂ue ∂ 1 ∂ e 2 ue = |ue | |Eh |2 = ∂x 4 ∂x 2ωpe mpe ∂x ∂ 2 |E| . = (16πme n0 )−1 ∂x
(10.7.72)
This represents the fast time-averaged low-frequency ponderomotive force generated by the self-interaction of the high-frequency part of the electric field. This force moves the electrons out of the plasma regions where the electric field has a local maximum. In view of (10.7.72), equation (10.7.71) takes the form 2 ∂ 2 n 1 ∂2 2 2 ∂ n |E| . − c = s ∂t2 ∂x2 16πmi ∂x2
(10.7.73)
This equation describes the formulation of density cavities due to the ponderomotive force related to the electric field. It is convenient to use the nondimensional quantities defined by x∗ =
x , λD
t∗ = ωpe t,
n∗ =
n , n0
E and E ∗ = √ . 4 πn0 KTe
In terms of these dimensionless quantities, dropping the asterisks, equations (10.7.66) and (10.7.73) represent Zakharov’s equations 1 ∂E 3 ∂2E = n E, + ∂t 2 ∂x2 2 ∂ 2 n ∂ 2 n 1 ∂2 2 |E| , − = ∂t2 ∂x2 4 ∂x2
iε
(10.7.74) (10.7.75)
1
e 2 where ε = ( m mi ) . We seek a stationary wave solution of this system as
n ,
E ∼ φ(x − M t) = φ(ξ),
so that equation (10.7.75) reduces to
4n 1 − M 2 = −|E|2 .
(10.7.76)
(10.7.77)
Using (10.7.77), equation (10.7.74) gives the nonlinear Schrödinger equation 3 iεEt + Exx + γ|E|2 E = 0, 2
(10.7.78)
εM x E(x, t) = ψ(ξ) exp i − ωt 3
(10.7.79)
where γ = 18 (1 − M 2 )−1 . We write
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics
in (10.7.78) so that ψ satisfies the equation 2 M2 2 − εω ψ + γψ 3 = 0. ψ − 3 6 3
563
(10.7.80)
This nonlinear equation admits a solution in the form 12 12
2 1 1 M2 2 M ψ(ξ) = − εω − εω sech ξ , 8 2γ 6 3 6
(10.7.81)
so that (10.7.79) gives the explicit representation of E(x, t). Physically, the solution represents envelope solitons only at subsonic speeds (M < 1). These envelope solitons travel faster, but become smaller and narrower as M increases. Schmidt (1975) investigated the instability and collapse of Langmuir solitons in some detail and found that the envelope solitons of Zakharov’s equations are stable to onedimensional perturbations, but unstable to two-dimensional perturbations. Considerable research on envelope solitons was done by several authors including Karpman (1971, 1975a, 1975b), Gibbons et al. (1977), Rowland et al. (1981), and Infeld and Rowlands (1990). Example 10.7.5 (Motion of an Electron Fluid). The basic equations of motion for the one-dimensional adiabatic motion of an electron fluid are ∂ ∂n + (nu) = 0, ∂t ∂x ∂u ∂u 1 ∂ e ∂φ +u + (nT ) − = 0, ∂t ∂x mn ∂x m ∂x d −2 n T = 0, dt ∂2φ − 2 + 4πe(n − n0 ) = 0, ∂x
(10.7.82) (10.7.83) (10.7.84) (10.7.85)
where n and u are the density and the flow velocity, respectively, φ is the electrostatic potential, and T is the electron temperature. Eliminating T from (10.7.83), (10.7.84) gives 3T0 ∂u e ∂φ ∂u ∂n +u + − = 0, (10.7.86) n ∂t ∂x mn20 ∂x m ∂x where n0 and T0 are the average values of n and T , respectively. We introduce the transformation of variables ξ = ε(x − Ct),
τ = ε2 t,
and
θ = kx − ωt,
(10.7.87)
where ε is a small parameter and C is a constant to be determined. We expand n, u, and φ in powers of ε as n = n0 + εn(1) (ξ, τ, θ) + ε2 n(2) (ξ, τ, θ) + · · · .
(10.7.88)
564
10 The Nonlinear Schrödinger Equation and Solitary Waves
Substituting φ(1) = ψ(ξ, τ ) exp(iθ) + c.c. from the first-order equation, we obtain the dispersion relation 3T0 2 ω 2 = ωp2 + k , m
(10.7.89)
(10.7.90)
1
where ωp = (4πe2 n0 /m) 2 . The compatibility condition of the second-order equation gives 3T0 k ∂ω = . C= ∂k mω
(10.7.91)
Similarly, the compatibility condition of the third-order equation yields i
2 e ∂ψ 3 T0 ωp2 ∂ 2 ψ T0 k 6 − + 16ω 2 − ωp2 |ψ|2 ψ 3 2 6 ∂τ 2 mω ∂ξ m mωωp ω + R(ξ, τ )ψ = 0, (10.7.92) n0
where R is an arbitrary function of ξ and τ . If R is independent of ξ, equation (10.7.92) reduces to the form i
∂ψ ∂ 2ψ + α 2 + β|ψ|2 ψ + iγψ = 0, ∂ζ ∂η
(10.7.93)
where η = ε(ξ − Ct) = εξ − Cτ , ζ = ε2 t = ετ , and α, β, and γ are known constants given by Sanuki et al. (1972). In this case, αβ < 0, and γ is a damping coefficient. Equation (10.7.93) is called the generalized nonlinear Schrödinger equation since it is modified by including the Landau damping. Sanuki et al. (1972) derived the equation (10.7.93) based on the Vlasov equation to describe the slow, nonlinear modulation of the amplitude of plasma waves including the effects of Landau damping. Example 10.7.6 (Nonlinear Quasi-Harmonic Waves and Modulational Instability). We derive equations for quasi-harmonic waves in a nonlinear medium from the nonlinear Klein–Gordon equation utt − c2 uxx + d2 u + σu3 = 0,
(10.7.94)
where c, d, and σ are constants. We seek a quasi-harmonic wave solution of (10.7.94) with a slowly varying complex amplitude A(x, t) in the form u(x, t) = A(x, t) exp i(ω0 t − k0 x) + c.c., (10.7.95) with frequency ω0 and wave number k0 chosen so as to satisfy the dispersion relation of the linear equation (10.7.94) with σ = 0
10.7 Examples of Physical Applications in Fluid Dynamics and Plasma Physics
ω 2 = d2 + c2 k 2 .
565
(10.7.96)
We next calculate the second-order derivatives and the cubic term in (10.7.94) so that
utt = Att + 2iω0 At − ω02 A eiθ0 + c.c.,
uxx = Axx − 2ik0 Ax − k02 A eiθ0 + c.c., u3 = A3 e3iθ0 + A∗3 e−3iθ0 + 3A|A|2 eiθ0 + 3A∗ |A|2 e−iθ0 , where θ0 = ω0 t − k0 x. Substituting these expressions into (10.7.94) and equating the coefficients of exp(iθ0 ) yields an equation for the complex amplitude A(x, t) i(At + cg Ax ) +
1 Att − c2 Axx + α|A|2 A = 0, 2ω0
(10.7.97)
3σ where cg = c2 (k0 /ω0 ) = ( dω ) is the group velocity and α = ( 2ω ) is the dk k=k0 0 coefficient of cubic nonlinear term. We use the variables ξ = x − cg t, τ = t in (10.7.97) so that the wave envelope varies slowly in the reference frame moving with the group velocity and equation (10.7.97) reduces to the nonlinear Schrödinger equation
iAτ − βAξξ + α|A|2 A = 0,
(10.7.98)
2
where β = c2g − c2 = 12 ( ddkω2 ) represents the dispersion parameter. We next use the transformation A = a exp(iφ) to obtain the system of equations for the real variables a = a(ξ, t) and φ = φ(ξ, t) in the form φt − βφ2ξ + βa−1 aξξ = αa2 , at − β(2aξ φξ + aφξξ ) = 0.
(10.7.99) (10.7.100)
We next use a new variable U = −2βφξ proportional to the variation of the wave number. This leads to the system of equations ∂ 1 (10.7.101) Ut + U Uξ = −q(a2 )ξ + 2β 2 aξξ , ∂ξ a (a2 )t + (a2 U )ξ = 0,
(10.7.102)
2
where q = 2αβ = α( ddkω2 ) is the product of the dispersion and nonlinear parameters. The instantaneous frequency ω and local wavenumber k of the modulated wave (10.7.95) are ω = ω0 + φt ,
k = k0 − φξ = k0 +
1 U (ξ, t). 2β
(10.7.103)
It follows from (10.7.98)–(10.7.102) that the properties of nonlinear envelope waves are described by the relation between the dispersion parameter β and the nonlinear parameter α. We can use the analysis of nonlinear geometrical optics to study
566
10 The Nonlinear Schrödinger Equation and Solitary Waves
nonlinear envelope waves. We consider only slow modulation so that the term with the third derivative corresponding to the group dispersion may be neglected. Consequently, the modulation equations (10.7.101)–(10.7.102) assume the hydrodynamic form
2
a t + U a2 ξ = 0. (10.7.104) Ut + U Uξ = −q a2 ξ , This reveals that, unlike linear theory, the frequency modulation depends on the variation of amplitude even in the geometrical approximation. We now investigate the modulational instability, that is, the behavior of small perturbations of the amplitude and the frequency of a stationary harmonic wave specified by the constant values a = a0 and U = U0 = 0. We next examine the instability of the monochromatic wave by linearizing the system (10.7.103) in the neighborhood of stationary values and then putting a = a0 + a (ξ, τ ) and U = U (ξ, τ ) so that a /a0 1 and U 1. These allow us to obtain a linear system 1 aτ + a0 Uξ = 0. (10.7.105) 2 We seek solutions of (10.7.104) in the form of harmonic waves a , U ∼ exp[i(Ωτ − κξ)] for the real wave number κ. This leads to the dispersion relation of the wave envelope in the form √ (10.7.106) Ω 2 = qa20 κ2 , or Ω = ± q(a0 κ). Uτ + 2qa0 aξ = 0,
Thus, if q > 0, the frequency Ω has two real values, that is, the system (10.7.103) is hyperbolic and the velocity of wave propagation takes two possible real values √ ± qa0 . Consequently, the small perturbation may be represented as a superposition of slow and fast waves, and the wave is stable in this case. On the other hand, if q < 0, the system (10.7.103) is elliptic and the frequencies are imaginary, and any small initial perturbation will grow in time as exp( |q|a0 τ ) → ∞ as τ → ∞, that is, a small perturbation becomes unstable and self-modulation occurs. Physically, the effect of the self-modulation is described by the growth of the sideband components of the wave spectrum into which the energy is transferred from the main component. Thus, the modulational instability occurs provided the nonlinear parameter α and the dispersion parameter β have opposite signs, that is, this product d2 ω < 0. (10.7.107) dk 2 This is known as the Lighthill Criterion for modulational instability, and it means that the dispersive and nonlinear effects compensate each other so that the spectral components corresponding to modulation interact synchronously with the carrier wave and grow. There are many physical examples of self-modulation. For example, self-modulation occurs for nonlinear Stokes waves on the surface of deep water, and eventually leads to their decomposition into wave packets. Such phenomena have been observed in a plasma and in electromagnetic waves. The modulational instability also occurs for flexural waves in thin rods and plates. q = 2αβ = α
10.8 Applications to Nonlinear Optics
567
10.8 Applications to Nonlinear Optics Nonlinear optics deals with the study of how high intensity light interacts with and propagates through matter. The discovery of laser (light amplification by stimulated emission of radiation) in 1960 led to a remarkably exciting development in the field of nonlinear optics. When a strong laser beam propagates in a medium, it may happen that the refractive index of the medium may change due to various effects produced by the interaction of the beam with the medium. In fact, it is found that the refractive index n is a function of the intensity of the light. One of the most interesting topics in nonlinear optics is the self-focusing of beams which can be described by the nonlinear Schrödinger equation. Evidently, this equation and its solitary wave solution play a fundamental role in this new, rapidly growing field of nonlinear optics. We start with the classical theory for the propagation of light based on the Maxwell equations which, for the first time, combine a unified treatment of electric and magnetic fields. The Maxwell equations for the electric and magnetic fields E and H and the electric and magnetic induction fields D and B are ∂B , ∂t ∂D ∇×H = + j, ∂t ∇ · D = ρ, ∇ · B = 0, ∇×E = −
(10.8.1a) (10.8.1b) (10.8.2a) (10.8.2b)
where ρ and j are electric charge and current densities, respectively. The law of conservation of charge is associated with the Maxwell equations. Adding the time derivative of (10.8.2a) with the divergence of (10.8.1b) gives ∂ρ + ∇ · j = 0. ∂t Integrating this equation over an entire volume V yields ∂ ρ dV = − ∇ · j dV = − j · n dS. ∂t V V S
(10.8.3)
(10.8.4)
This expresses the fact that an increase or decrease of charge in a given volume V is affected by the transport of charge, that is, the flow of current, through the boundaries of V . Since we deal with dielectrics in which there are no free charges and in which no current flows, we get ρ = 0 and j = 0. Equations (10.8.1a), (10.8.1b), (10.8.2a), (10.8.2b) combined with constitutive relations connecting B, H, D, and E represent the field equations. The constitutive equations are B = μH,
D = ε0 E + P = εE,
(10.8.5ab)
where μ = (ε0 c−2 ) and ε (both of which are constants) are called the magnetic permeability and dielectric constants, respectively, and c is the speed of light in a
568
10 The Nonlinear Schrödinger Equation and Solitary Waves
vacuum. The vector P is called the polarization, and it is zero in a vacuum. When light propagates in a dielectric medium, the electric field causes distortion in the atomic structure, generating local dipole moments, and thereby induces a polarization field P which depends on the electric field E. For small to moderate values of the electric field amplitude and if there is no resonance between the electric field and the medium, P is a linear function of E. In an isotropic medium which responds instantaneously to the electric field, P = ε0 χE, where χ is called the electric susceptibility. Taking the curl of (10.8.1a) and utilizing (10.8.1b) and (10.8.2a) gives ∇2 E −
2 −1 ∂ 2 P 1 ∂2E − ∇(∇ · E) = ε0 c . c2 ∂t2 ∂t2
(10.8.6)
This represents the propagation of the electric field, valid for both linear and nonlinear wave propagation. Since P = ε0 χE, and χ is a constant, (10.8.6) reduces to the wave equation 2 2 n ∂ E 2 ∇ E− = 0, (10.8.7) c ∂t2 √ where n = 1 + χ(> 1) is a nondimensional quantity called the refractive index of the medium. Equation (10.8.7) shows that the electric field E propagates with the velocity (c/n). We seek singly polarized, linear plane wave solutions of equation (10.8.6) in the form E(x, t) = e A exp(iθ) + c.c. , (10.8.8) where e is a unit vector in the direction of polarization of E, A is a complex amplitude, θ = k · x − ωt, k = (k, , m), and c.c. stands for the complex conjugate. These waves are called monochromatic waves because they contain only one wavevector k and one frequency ω. Such solutions (10.8.8) exist for A = 0, provided that the dispersion relation D(k, ω)A =
k2 + 2 + m2 −
nω c
2
A=0
(10.8.9)
is satisfied. For nonlinear dispersive media, D also depends on the wave amplitude. In general, D = 0 can be solved for any one of the variables, k, , m, and ω in terms of the other three. In optics, it is customary to solve for the wavenumber in terms of frequency because the refractive index n depends on the frequency, that is, n = n(ω). For linear or nonlinear waves, the dispersion equation (10.8.9) is algebraic. For wavepackets in a linear dispersive media, (10.8.9) is a linear partial differential equation for A that describes how the wavepacket envelope A moves with the group velocity and is modified by dispersion and diffraction. However, for wavepackets in a nonlinear dispersive medium, (10.8.9) represents a nonlinear partial differential equation for the complex wave envelope A. For the simplest nontrivial case, A satisfies the nonlinear Schrödinger equation, which is a very basic equation in nonlinear
10.8 Applications to Nonlinear Optics
569
optics. Moreover, the dispersion relation is one of the most fundamental concepts in the theory of wave propagation. The dispersion relation (10.8.9) is also valid when the refractive index of the n ) nλ , where L is the distance over which medium varies slowly, that is, ∇n = O( L n changes. We assume that the electric susceptibility χ, and hence, the refractive index n are slowly varying functions of the space variable x, and we also suppose that E = yE is polarized in the y-direction. This implies that equation (10.8.7) holds for the scalar field E(x, z, t). We seek a solution E(x, z, t) = A exp(iθ) + c.c. + εE1 ,
(10.8.10)
where k = (k, m) = ∇θ and ω = −θt denote the wavevector and frequency of the λ
1. electromagnetic wave, and ε = L λ Since n varies slowly with x, that is, n = n(X = εx), where ε = L
1, hence, k and ω also vary slowly. If ω is assumed to be constant, and since k = θx and m = θz , then kz = mx holds. Since k is independent of z, we require that m also be constant, and hence, only the wavenumber k can change in the direction of variable refractive index. Substituting the solution (10.8.10) in equation (10.8.7) and assuming that n, A, and k are slowly varying functions of x, that is, functions of X, gives the dispersion relation (10.8.9) to leading order
n2 (X)ω 2
2 2 D(k, 0, m, ω)A = k + m − A = 0. (10.8.11) c2 We also find that O(ε) : ∇2 E1 −
2 2 ∂A n ∂ E dk = −2ik − i A eiθ + c.c. c ∂t2 ∂X dX
(10.8.12)
To eliminate solutions of E1 that grow like z exp(±iθ), we must set 2k or
dA dk + · A = 0, dX dX
(10.8.13)
kA2 = const. = k0 A20 .
(10.8.14)
Therefore, k(x) and A(x)are given by
12 ω2 k(x) = 2 n2 (X) 2 − m2 , c
A(x) =
k0 k
12 A0 ,
(10.8.15ab)
provided that k = 0. Equation (10.8.14) is called the eikonal equation, and it represents the conservation of wave action. This method of analysis was first proposed by Wentzel, Kramers, Brillouin, and Jeffrey and is known as the WKBJ method. Initially, with n0 ω ( nω c ) > m, the wave cannot propagate, and k > 0. However, if ( c ) < m, there ω exists a point x, called a caustic, at which n(x1 ) · c = m and k = 0. At this
570
10 The Nonlinear Schrödinger Equation and Solitary Waves
point, the light rays turn around and the electromagnetic wave is reflected back to )2 − m2 ]. The exact nature near x = x1 cannot be the medium with k = − [( nω c determined by the WKBJ method because k → 0, the amplitude A increases like 1 O(k − 2 ), and the WKBJ method fails. In this case, we can approximate n2 locally by {n2 (x1 ) − a(x − x1 )} and seek a solution of the form E(x, t) = B(x) exp i(kz − ωt) + c.c.,
k=
ωn0 . c
(10.8.16)
It turns out that B(x) satisfies the Airy differential equation Bxx −
2 ω a(x − x1 )B = 0. c
(10.8.17)
When x > x1 , we obtain a solution of (10.8.17) which decays exponentially. On the other hand, when x < x1 , the solution of (10.8.17) has two components: one corresponds to an incoming wave with the wavenumber (−k, 0, m), where k = 2 2 12 −{( nω c ) −m } and the amplitude is given by (10.8.15ab), and the other represents the reflected wave with the wavenumber (k, 0, m). This reveals the fact that light rays turn toward regions of higher refractive index and turn away from regions of lower refractive index. This fact is widely used in the theory of the propagation of light in dielectric materials and has tremendous ramifications. First, the correction to the refractive index is an increasing function of the field intensity, that is, n2 2 n = n0 + n2 |A|2 = n0 1 + |A| , n2 > 0. (10.8.18) n0 If the medium is linear, then n2 = 0. The electromagnetic waves are governed by the standard linear wave equation. So the quantity δ = nn20 |A|2 measures the strength of nonlinearity, and produces what is called the Kerr effect. When light propagates in a Kerr dielectric medium, the light continues to focus in regions of greatest intensity. This instability is called the self-focusing instability in optics, and it is widespread in nature. In certain cases, a wavetrain of amplitude which is uniform in space and time becomes unstable, and light intensity concentrates in collapsing filaments. If there is only one transverse direction, a balance between diffraction and nonlinearity leads to the formation of solitons. Hasegawa and Tappert (1973) proved, mathematically, that an optical pulse in a dielectric fiber forms an envelope soliton, that is, the pulses of light either do not change shape or change shape periodically. Mollenauer et al. (1986) demonstrated this effect experimentally. An interesting discussion of these developments has been given by Hasegawa (1990). In two transverse dimensions, the collapse is no longer confined, and the light filament continues to intensify. In reality, light waves are not exactly monochromatic. They usually consist of a linear combination of plane waves whose wavevectors and frequencies are confined to a narrow band of order ε about some central values k0 and E0 , where 0 < ε ≤ 1. Such plane waves are called wavepackets represented in the form (10.8.19) E= e A(x, t) exp i(k0 · x − ωt) + c.c. + O(ε),
10.8 Applications to Nonlinear Optics
571
combined with the dispersion relation D(k0 , 0 , m0 , ω0 ) = 0,
(10.8.20)
where the wave amplitude A(x, t) is a slowly varying function of space and time, that is, ∂A (10.8.21) = ε|k0 |2 A |k0 |A. ∂x We introduce the notions of local wavevector k and local frequency ω, and write the complex amplitude in the polar form A = a exp(iφ), where a and φ are real. Then the phase of the leading order contribution to the electric field is given by θ = k0 · x − ω0 t + φ(x, t),
(10.8.22)
so that k = ∇θ = k0 + ∇φ
and
ω = −θt = ω0 − φt .
(10.8.23ab)
Evidently, ∂k + ∇ω = 0. (10.8.24) ∂t This represents the conservation of wavenumbers. We consider the evolution of the wavepackets (10.8.19), with k0 = (0, 0, m0 ) and e = (0, 1, 0) of an electric field E that satisfies equation (10.8.7). For simplicity, we first consider the case of constant n. Substituting (10.8.19) in (10.8.7) gives 2 2 n ∂ A ∂A ∂ 2 A ∂ 2 A ∂A + − + + 2iω0 = 0, (10.8.25) 2im0 2 2 ∂z ∂z ∂x ∂t c ∂t2 where m0 = ( nωc 0 ). Since each derivative is of order ε, the dominant terms in (10.8.25) are the first and the fourth, so, to the leading order, (10.8.25) yields n Az + At = 0. (10.8.26) c Thus, to a first-order approximation, A is a function of {t − ( nc )z} and is constant when {t − ( nc )z} is constant, that is, in a frame of reference moving with the group velocity c . (10.8.27) Cg = ∇k ω = 0, 0, n In view of (10.8.26), it turns out that Azz = ( nc )2 Att , and so, to order ε3 , equation (10.8.25) becomes 2 ∂A ∂ A n ∂A i = 0. (10.8.28) + − ∂z c ∂t 2m0 ∂x2 This is usually known as the paraxial equation because the rays, the normals to the planes of constant phase, of all the plane waves in the wavepacket are nearly
572
10 The Nonlinear Schrödinger Equation and Solitary Waves
parallel. Invoking the change of variables to a frame of reference moving with the group velocity, z = z, τ = {t − ( nc )z}, equation (10.8.28) assumes the form i Az − Axx = 0. (10.8.29) 2m0 This equation describes the dispersion of a wavepacket in the transverse direction. Such a phenomenon in optics is called wave diffraction. We next recall the dispersion relation (10.8.9) with constant refractive index which was originally derived by using A exp{i(k · x − ωt)} in (10.8.7). If, instead of substituting a plane wave for E, we insert the wavepacket (10.8.19) with k0 = (0, 0, m0 ), we obtain a modified dispersion relation, ∂ ∂ ∂ ∂ A(x, y, z, t) = 0. (10.8.30) D −i , −i , m0 − i , ω0 + i ∂x ∂y ∂z ∂t This is an equation for the slowly varying wave envelope A. We now utilize the frequency dependence of the refractive index, and use D = (k2 + 2 + m2 ) − { n(ω)ω }2 , so that equation (10.8.30) becomes c 2 2 2 ∂ ∂ ∂ −i + −i + m0 − i ∂x ∂y ∂z 2
1 ∂ ∂ − 2 n ω0 + i A(x, t) = 0. ω0 + i c ∂t ∂t
(10.8.31)
To first order, we recover the dispersion relation for the plane waves m20 = n2 (ω0 )
ω02 . c2
To the order of ε, we obtain the equation
∂A ∂ ωn(ω) ∂A + = 0. ∂z ∂ω c ω=ω0 ∂t
(10.8.32)
(10.8.33)
This is consistent with the fact that, to the leading order, the wave envelope A of a wavepacket propagates with the group velocity Cg , where Cg−1 = ( ∂m ∂ω )0 . Next, we expand (10.8.30) to order ε2 and retain all terms up to second derivative to obtain the equation 2i ∂ (2im0 )Az + Axx + ω (nω) n(ω ) At + Azz 0 0 c2 ∂ω 0 2
∂ 1 ∂ Att = 0. (10.8.34) (nω) + (n0 ω0 ) − 2 (nω) c ∂ω ∂ω 2 0 0
10.8 Applications to Nonlinear Optics
573
Using (10.8.33), we can simplify (10.8.34) in the form 2 2
2m 1 ∂2m ∂ m ∂ m At + Axx + Ayy + Att = 0. Az + 2 2 2ω 0 2 ∂k ∂ ∂ω 2 0 0 0 (10.8.35) We now incorporate the effects of small variations of the refractive index to the order ε2 in (10.8.35) by replacing n in (10.8.31) by (n + δn). Consequently, an extra term c−2 (2nω02 δn)A will be included in equation (10.8.34), and, if we divide across by (2im0 ), an additional term −iδn( ωc0 )A = −i( δn n )m0 A is generated in (10.8.35), ∂m and this term becomes −i( ∂n )0 δnA. Thus, equation (10.8.35) reduces to the form 2 2 2 ∂ m i ∂m ∂ m ∂ m At + Axx + Ayy + Att Az + 2 2 ∂ω 0 2 ∂k 0 ∂ ∂ω 2 0 0
∂m (δn) A = 0. (10.8.36) −i ∂n
This equation is widely used in nonlinear optics. Finally, we derive the nonlinear Schrödinger equation for the evolution of a wavepacket envelope A(x, t) in a dispersive medium with a nonlinear refractive index n given by (10.8.18). The nonlinear dispersion relation is then given by
D k, , m, ω, |A|2 A =
k 2 + 2 + m2 −
nω c
2 A = 0.
(10.8.37)
For a wave amplitude A of order ε, so that Azz = O(ε3 ), |A|2 A = O(ε3 ), and hence, the variation of δn in refractive index is of order ε2 , the resulting equation for the wave envelope A is the same as (10.8.36) with δn = n2 |A|2 . To include the effects of slow variations of space variables and amplitude, we replace δn in (10.8.36) by δn(x, y) + n2 |A|2 , so that dropping the subscripts zero on m and the coefficients, equation (10.8.36) becomes 2 2
2 i ∂m ∂ m ∂ m ∂ m + + Az + A A Att At + xx yy ∂ω 2 ∂k2 ∂2 ∂ω 2 im − δn + n2 |A|2 A = 0. (10.8.38) n Or equivalently, i ∂m im At + (αAxx +βAyy +γAtt )− δn+n2 |A|2 A = 0, (10.8.39) Az + ∂ω 2 n where the three coefficients of dispersive terms are α=
∂ 2m , ∂k 2
β=
∂2m , ∂2
and
γ=
∂2m . ∂ω 2
(10.8.40)
574
10 The Nonlinear Schrödinger Equation and Solitary Waves
If the associated carrier wave is proportional to exp{i(k · x − ωt)}, where k = (0, 0, m)
and m =
n(ω)n , c
(10.8.41)
then
∂m 1 ∂ 1 = (nω) and α = β = − . (10.8.42) ∂ω c ∂ω m Thus, the resulting nonlinear Schrödinger equation becomes 2 ∂ i 1 ∂ i (nω) At − (Axx + Ayy ) + (nω) Att Az + c ∂ω 2m 2c ∂ω 2 im δn + n2 |A|2 A = 0. (10.8.43) − n
The above analysis reveals that the NLS equation (10.8.39) or (10.8.43) describing the self-focusing of light beams is the fundamental equation in nonlinear optics. In particular, if A is independent of x, y, and t, and δn ≡ 0, the solution of the resulting equation (10.8.39) is given by
n2 A(z) = A0 exp im |A0 |2 z . (10.8.44) n This solution can be used to determine the wavenumber of the electric field E = E[A(z) exp{i(mz − ωt)} + c.c.]. The value of this wavenumber is equal to m(1 + nn2 |A|2 ) or ωc (n + n2 |A|2 ). According to the theory of the NLS equation, it also reveals that, if the product of (mn2 ) with any one of three dispersive coeffi1 1 and (mn2 )(− m ) = −n2 ) is negative, then cients α, β, or γ (that is, α = β = − m the amplitude-modulated solution (10.8.44) is definitely unstable. In particular, if the wavepacket envelope A is independent of t with α = β = 1 −m , equation (10.8.39) becomes i im δn + n2 |A|2 A = 0. (Axx + Ayy ) − (10.8.45) Az − 2m n Further, if we assume that δn = 0, this equation reduces to the form 1 m iAz + n2 |A|2 A = 0. (Axx + Ayy ) + 2m n
(10.8.46)
If, initially, the field envelope A = A0 (z) is independent of x and y, then a small perturbation δA confined at some point causes the refractive index n2 |A|2 to increase there. The light rays from the neighboring regions continue to focus toward this region, enhancing the intensity more and more there. As more light focuses there, the refractive index becomes even higher, and even more light converges. Thus, the effect of an increased refractive index leads to focusing instability in nonlinear optics. However, this instability is not confined to transverse modulations. We now
10.9 Exercises
575
derive the nonlinear Schrödinger equation for a concentrated wave mode traveling in a waveguide, and, for this model, there is no transverse dispersion but only dispersion in the direction of the wave propagation. So, if we suppose that A(z, t) is independent of x and y with the fact that δn = 0, then (10.8.39) becomes ∂m n2 i ∂2m At + |A|2 A = 0. A Az + (10.8.47) tt − im ∂ω 2 ∂ω 2 n 2
Obviously, the focusing instability occurs provided that the product ( ∂∂ωm2 ) × mn2 2 is positive. For the diffraction case, the product is ( nn2 ) whose (− n ) = − γmn n sign depends on that of n2 . In this case of wave propagation in an optical fiber, it can be shown that n2 > 0 only if γ < 0. However, in general, γ can be positive or negative in the neighborhood of resonance where the real part of the susceptibility depends on frequency. These situations are referred to as normal dispersion and anomalous dispersion, respectively. According to the general theory of the NLS equation, there exists a stable soliton solution of (10.8.47) in nonlinear optical fibers. Two solitons of different frequencies have different group velocities and pass through each other without distortion. However, if two solitons have the same frequency, and hence, the same velocity, they undergo a periodic oscillation. There is a remarkable change in the properties of solutions of (10.8.39) when two transverse dimensions and time are included. In the defocusing case n2 < 0 (n2 γ > 0), uniform wavetrain solutions are neutrally stable. In fact, deformations in the uniform wavetrains, in which intensity diminishes locally, neither grow nor decay, but evolve as solitons of a different nature. These solitons are called dark solitons because they have intensities lower than those of the ambient wavetrains. We close this section by adding some brief comments on recent applications of nonlinear optics to modern technology. The use of lasers ranges from high-density data storage on optical disks to greatly improved surgical techniques in ophthalmology, neurosurgery, and dermatology. The discovery of optical fibers led to a revolutionary growth in modern telecommunications. There has been a tremendous impact of optical fibers on the fundamental study of nonlinear optical interactions. Once high-quality, low-loss, single-mode fibers became available in the 1980s, the number of possible new applications has grown in a remarkable fashion. Soliton lasers, modulational instability lasers, SBS lasers, and various kinds of optical devices were built, and the open cavity resonator marked the critical breakthrough in the design of working lasers. It is clear that nonlinear optics is likely to revolutionize future telecommunications and computer technology.
10.9 Exercises 1. (a) Show that the dispersion relation (7.2.11b) for the Boussinesq equation (7.2.11a) in the long-wave approximation (k → 0) is given by 1 2 2 ω = ω(k) = c0 k 1 − β k . 2
576
10 The Nonlinear Schrödinger Equation and Solitary Waves
(b) Hence, show that the Fourier integral representation around k0 = 0 is approximately equal to
1 F (δk) exp i δk(x − ct) + cβ 2 (δk)3 t d(δk). φ(x, t) ∼ 2 δk (c) Introduce the slow variables ξ = ε(x − ct) and τ = ε3 t, where δk = 0(ε), and then show that the long wave field φ(x, t) = A(ξ, τ ) satisfies the linearized KdV equation 1 Aτ + cβ 2 Aξξξ = 0. 2 2. Show that the NLS equation iut + uxx + |u|2 u = 0 has the combined rational and oscillatory (Peregrine 1983) solution
u(x, t) = 1 − 4(1 + 2it)/ 1 + 2x2 + 4t2 exp(it). 3. Show that the NLS equation in Exercise 2 has the Ma solitary wave solution (Ma 1979; Peregrine 1983) of the form
u(x, t) = a exp ia2 t 1 + 2m(m cos θ + in sin θ)/f (x, t) for all real√a and m, where n2 = (1 + m2 ), θ = 2mna2 t, and f (x, t) = n cosh(ma 2x) + cos θ. 4. Seek a traveling wave solution of the NLS equation in Exercise 2 in the form u(x, t) = r exp i(θ + ωt) , where r(x − ct) and θ(x − ct) are real functions and c and ω are real constants. Prove that 1 A θ = c+ and (R )2 = −2F (R), 2 R where R = r2 , F (R) = R3 − 2(ω − 14 c2 )R2 + BR + A, and A and B are arbitrary real constants of integration. Hence, show that there exists a solitary wave solution of the form
a 1 u(x, t) = a exp i c(x − ct) + ωt sech √ (x − ct) , 2 2 for all a2 such that
1 a2 = 2 ω − c2 > 0. 4 5. Following the same method as given in Exercise 4, show that there exists a solitary wave solution (Zakharov and Shabat 1973) of the NLS equation iut + uxx − |u|2 u = 0 with r2 (ξ) = m − 2κ2 sech2 (κξ) and c tan{θ(ξ)} = −2κ tanh(κξ) for all c, 1 where ξ = x − ct, ω = −m, and κ = 12 (2m − c2 ) 2 > 0.
10.9 Exercises
577
6. Show that an auto-Bäcklund transformation (Konno and Wadati 1975) for the NLS equation iut + uxx + 2|u|2 u = 0 is the following pair of equations: 1 ux + vx = (u − v) 4λ2 − |u + v|2 2 , 1 ut + vt = i(ux − vx ) 4λ2 − |u + v|2 2 1 + i(u + v) |u + v|2 + |u − v|2 , 2 where u and v satisfy the equation. Hence, find the solitary wave solution of this NLS equation by setting v = 0 for all x and t. 7. Derive the NLS equation for the local amplitude A(x, y, t) from the resonant interaction equations (Phillips 1981; Debnath 1994) in the form ω0 ω0 ω0 Ax + i Axx − i Ayy = b˙ exp(iθ) dk. At + i 2k0 8k02 4k02 2
1
8. If m = [{n2 (ω)( ωc2 )} − (k 2 + 2 )] 2 , show that ∂2m 1 ∂m ∂m ∂ 2m = =− , = = 0, ∂k ∂ ∂k 2 ∂2 m0
∂ nω ∂m ∂2m ∂ 2 nω = , = , ∂ω ∂ω c ∂ω 2 ∂ω 2 c 0 0 where the derivatives are computed at k = = 0, m = m0 = n(ω0 )( ωc0 ). 9. Show that the solution
A(z) = A0 exp iγ|A0 |2 z of the nonlinear Schrödinger equation Az − iα(Axx + Ayy ) = iγA2 A∗ is unstable. Also show that the maximum growth rate occurs for k = |A0 | γ/α. 10. Obtain the solution of the De Broglie wave problem for an electron of mass m in a constant electric field E along the x-axis. The motion of the electron is described by the wave function ψ(x, t) that satisfies the Schrödinger equation
2 ∂ 2 ψ, iψt = V (x) − 2m ∂x2 where is the Planck constant and V (x) = xE is the potential energy of the electron.
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
As long as a branch of science affords an abundance of problems, it is full of life; want of problems means death or cessation of independent development. Just as every human enterprise prosecutes final aims, so mathematical research needs problems. Their solution steels the force of the investigator; thus he discovers new methods and view points and widens his horizon. David Hilbert The enormous usefulness of mathematics in the natural sciences is something bordering on the mysterious and there is no rational explanation for it. It is not at all natural that “laws of nature” exist, much less that man is able to discover them. The miracle of the appropriateness of the language of mathematics for the formulation of the laws of physics is a wonderful gift which we neither understand nor deserve. Eugene Wigner
11.1 Introduction This chapter deals with the theory and applications of nonlinear Klein–Gordon (KG) and sine-Gordon (SG) equations. Special emphasis is given to various methods of solutions of these equations. The Green function method combined with integral transforms is employed to solve the linear Klein–Gordon equation. The Whitham averaging procedure and the Whitham averaged Lagrangian principle are used to discuss solutions of the nonlinear Klein–Gordon equation. Included are different ways of finding general and particular solutions of the sine-Gordon equation. Special attention is given to solitons, antisolitons, breather solutions and the energy associated with them, interaction of solitons, Bäcklund transformations, similarity solutions, and the inverse scattering method. Significant features of these methods and solutions are described with other ramifications. L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_11,
580
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
11.2 The One-Dimensional Linear Klein–Gordon Equation The one-dimensional, inhomogeneous, Klein–Gordon equation is given by utt − c2 uxx + d2 u = p(x, t),
x ∈ R, t > 0
(11.2.1)
with the initial boundary conditions u(x, 0) = 0 = ut (x, 0) u(x, t) → 0
for x ∈ R,
as |x| → ∞, t > 0,
(11.2.2ab) (11.2.3)
where c and d are constants. The Green function G(x, t) associated with this problem satisfies the equation Gtt − c2 Gxx + d2 G = δ(x)δ(t),
x ∈ R, t > 0,
(11.2.4)
with the same initial and boundary conditions. Application of the joint Laplace and Fourier transform (Debnath 1995) gives 1 1 ˜ G(k, s) = √ , (11.2.5) 2 + α2 ) (s 2π 1
where α = (c2 k 2 + d2 ) 2 . The inverse Laplace transform yields ˜ t) = √1 G(k, 2π
sin αt . α
(11.2.6)
Finally, the inverse Fourier transform leads to the solution − 12 ∞ d2 d2 1 1 2 2 sin ct k + 2 exp(ikx) dk k + 2 G(x, t) = c 2π −∞ c c
d 1 J0 = c2 t2 − x2 H ct − |x| 2c c 1 √ d J [ c2 t2 − x2 ] if |x| < ct, 2c 0 c = (11.2.7) 0 if |x| > ct. In the limit as d → 0, this result is in perfect agreement with the solution for the standard wave equation. If the source is located at (ξ, τ ), Green’s function assumes the form 1 1 J [ d {c2 (t − τ )2 − (x − ξ)2 } 2 ] if |x − ξ| < c(t − τ ), 2c 0 c G(x, t; ξ, τ ) = 0 if |x − ξ| > c(t − τ ). (11.2.8) In terms of the Heaviside functions, Green’s function can be expressed as
11.2 The One-Dimensional Linear Klein–Gordon Equation
G(x, t; ξ, τ ) =
581
1 d 2 1 J0 c (t − τ )2 − (x − ξ)2 2 H x + ct − (ξ + ct) 2c c × H ξ + ct − (x + cτ ) . (11.2.9)
In the limit as d → 0, this result reduces to that for the wave equation. The solution of the Klein–Gordon equation (11.2.1) with the initial conditions u(x, 0) = f (x),
ut (x, 0) = g(x),
x ∈ R,
(11.2.10)
can be expressed in terms of Green’s function as t ∞ p(ξ, τ )G(x, t; ξ, τ ) dξ dτ u(x, t) = 0 −∞ ∞ g(ξ)G(x, t; ξ, 0) − f (ξ)Gτ (x, t; ξ, 0) dξ. (11.2.11) + −∞
We note that, in the double integral in (11.2.11), G = 0 for τ > 1, and hence, the limit in the τ integral extends only up to t. Also, it follows from (11.2.8) that G is nonzero when |x−ξ| < c(t−τ ), which is equivalent to x−c(t−τ ) < ξ < x+c(t−τ ). Consequently, the double integral in (11.2.11) becomes t ∞ dτ p(ξ, τ )G(x, t; ξ, τ ) dξ −∞
0
=
1 2c
t 0
x+c(t−τ )
dτ
p(ξ, τ )J0 x−c(t−τ )
1 d 2 c (t − τ )2 − (x − ξ)2 2 dξ. (11.2.12) c
Similarly, since the product of the Heaviside functions vanishes outside the interval (x − ct, x + ct), we find that ∞ x+ct
1 1 d 2 2 g(ξ)G(x, t; ξ, 0) dξ = g(ξ)J0 c t − (x − ξ)2 2 dξ. 2c x−ct c −∞ (11.2.13) It follows from direct integration that d 2 2 td J0 [ c c t − (x − ξ)2 ] Gτ (x, t; ξ, τ ) τ =0 = − 2 c2 t2 − (x − ξ)2 × H(x + ct − ξ)H(ξ − x + ct) 1 d 2 2 2 c t − (x − ξ) + J0 2 c × δ(x + ct − ξ)H(ξ − x + ct)
+ δ(ξ − x + ct)H(x + ct − ξ) . In view of this result, the second term within the square bracket of (11.2.11) becomes
582
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
−
∞
f (ξ)Gτ (x, t; ξ, 0) dξ x+ct J1 [ dc c2 t2 − (x − ξ)2 ] td f (ξ) dξ =− 2 c2 t2 − (x − ξ)2 x−ct 1 ∞ d + J0 c2 t2 − (x − ξ)2 δ(x + ct − ξ)H(ξ − x + ct) 2 −∞ c
+ δ(ξ − x + ct)H(x + ct − ξ) f (ξ) dξ x+ct J1 [ dc c2 t2 − (x − ξ)2 ] td f (ξ) dξ =− 2 c2 t2 − (x − ξ)2 x−ct 1 (11.2.14) + f (x − ct) + f (x + ct) , 2 −∞
in which the property of the delta function with H(2ct) = 1 is used. Combining (11.2.12), (11.2.13), and (11.2.14), the final form of solution (11.2.11) is given by u(x, t) =
1 f (x − ct) + f (x + ct) 2 x+ct d 1 2 2 2 J0 c t − (x − ξ) g(ξ) dξ + 2c x−ct c x+ct J1 [ dc c2 t2 − (x − ξ)2 ] td − f (ξ) dξ 2 c2 t2 − (x − ξ)2 x−ct t x+c(t−τ ) d 1 dτ J0 c2 (t − τ )2 − (x − ξ)2 p(ξ, τ ) dξ. + 2c 0 c x−c(t−τ ) (11.2.15)
If d = 0, this solution reduces to that for the Cauchy problem for the inhomogeneous wave equation.
11.3 The Two-Dimensional Linear Klein–Gordon Equation The two-dimensional, linear, inhomogeneous, Klein–Gordon equation is utt − c2 (uxx + uyy ) + d2 u = p(x, y, t),
−∞ < x, y < ∞, t > 0.
(11.3.1)
The initial and boundary conditions are u(x, y, 0) = 0 = ut (x, y, 0) for all x and y, u(x, y, t) → 0 as r = x2 + y 2 → ∞, t > 0.
(11.3.2) (11.3.3)
The Green function G(x, y, t) for the two-dimensional Klein–Gordon equation satisfies the equation
11.3 The Two-Dimensional Linear Klein–Gordon Equation
Gtt − c2 (Gxx + Gyy ) + d2 G = δ(x)δ(y)δ(t),
583
−∞ < x, y < ∞, t > 0, (11.3.4)
with the conditions G(x, y, 0) = 0 = Gt (x, y, 0) for all x and y,
G(x, y, t) → 0 as x2 + y 2 → ∞, t > 0.
(11.3.5) (11.3.6)
In terms of polar coordinates (r, θ) without θ-dependence, equation (11.3.4) becomes 1 δ(r)δ(t) , r > 0, t > 0. (11.3.7) Gtt − c2 Grr + Gr + d2 G = r 2πr Application of the joint Laplace and Hankel transform of order zero gives 1 ˜ ¯ s) = 1 · G(κ, , 2π (s2 + c2 κ2 + d2 )
(11.3.8)
whence, the inverse Laplace transform yields ˜ t) = 1 · sin(αt) , G(κ, 2π α
1 α = c2 κ2 + d2 2 .
(11.3.9)
Finally, the inverse Hankel transform leads to the solution 1 G(r, t) = 2πc
∞ 0
d2 κ J0 (κr) κ + 2 c 2
− 12
1 d2 2 2 dκ, sin ct κ + 2 c (11.3.10)
which, by using the standard table of Hankel transforms, is √ 1 cos[( dc ) c2 t2 − r2 ] √ = H(ct − r) 2πc c 2 t2 − r 2 − 12 r r2 r2 1 2 2 . · t − 2 cos d t − 2 H t − = 2πc2 c c c
(11.3.11)
If the source is located at (ξ, η, τ ) instead of the origin, Green’s function assumes the form − 12 R 1 R2 R2 2 2 , G(x, y, t; ξ, η, τ ) = · T − 2 cos d T − 2 H T − 2πc2 c c c (11.3.12) where T = t − τ and R2 = (x − ξ)2 + (y − η)2 .
584
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
11.4 The Three-Dimensional Linear Klein–Gordon Equation The Green function G(x, t) for the three-dimensional Klein–Gordon equation satisfies the equation Gtt − c2 ∇2 G + d2 G = δ(x)δ(y)δ(z)δ(t),
−∞ < x, y, z < ∞, t > 0, (11.4.1)
with the conditions for all x, y, z, as |x| → ∞,
G(x, 0) = 0 = Gt (x, 0) G(x, 0) → 0
(11.4.2) (11.4.3)
where x = (x, y, z). The application of the joint Laplace and the triple Fourier transform (Debnath 1995) to this problem gives ˜ ¯ s) = G(κ,
1 1 · , (2π)3/2 (s2 + c2 κ2 + d2 )
where κ = (k, l, m). The joint inverse transform gives the solution ∞
sin αt 1 κ, κ · x) dκ exp i(κ G(x, t) = 3/2 α (2π) −∞
(11.4.4)
(11.4.5)
1
where α = (c2 κ2 + d2 ) 2 . Introducing spherical polar coordinates in the κ-integral and taking the polar axis along the direction of the vector x so that κ · x = κr cos θ, |x| = r, θ is the polar angle, and κ = κ2 dκ sin θ dθ dφ, we obtain ∞ π 2π 1 sin αt 2 κ G(x, t) = dφ dκ exp(iκr cos θ) sin θ dθ (2π)3 0 α 0 0 ∞ dκ 1 sin(αt) · κ sin(κr) = 2 (2π) r 0 α ∞ sin αt ∂ 1 =− · cos(κr) dκ. (11.4.6) (2π 2 r) 0 α ∂r ∂ can be written outside the Because integral (11.4.6) is an even function of κ, and ∂r integral, we now obtain − 12 ∞ ∂ d2 d2 1 2 exp(iκr) κ + 2 sin ct κ2 + 2 dκ G(x, t) = − 2 · 4π rc ∂r −∞ c c
1 ∂ d =− c2 t2 − r2 H c2 t2 − r 2 · J0 (11.4.7) 4πrc ∂r c
in which the standard table of the Fourier transforms is used.
11.5 The Nonlinear Klein–Gordon Equation and Averaging Techniques
585
Finally, we carry out the differentiation and use the identity J0 (kr) = −kJ1 (kr) and J0 (0) = 1 to obtain the Green function √
1 2 2 d J1 ( dc c2 t2 − r2 ) 2 2 √ δ c t − r2 − G(x, t) = H c t − r2 . 2π 2c c 2 t2 − r 2 (11.4.8) In the limit as d → 0, this result is in perfect agreement with that for the threedimensional wave equation.
11.5 The Nonlinear Klein–Gordon Equation and Averaging Techniques Whitham (1965a, 1965b) has successfully extended the averaging procedure for the study of nonlinear ordinary differential equations associated with oscillation problems to partial differential equations involved in nonlinear dispersive waves. The procedure that he developed was to average over local oscillations and then to obtain partial differential equations describing the slow variations of the basic wave parameters, such as wavenumber, frequency, and amplitude. In all problems where the equations admit locally uniform periodic wavetrains as solutions of the form u(x, t) = u(θ), where θ = κ · x − ωt is the phase function, u(θ) is a periodic function of θ, and k and ω are the wavenumber vector and frequency, respectively. In linear wave problems, u(x, t) = a cos θ, which is sinusoidal. According to the Whitham theory of slowly varying wavetrains, the solution of the form u = u(θ, a) is maintained, where a is not a constant and θ is not a linear function of x and t. The parameters ω, κ , and a are assumed to be slowly varying functions of x and t corresponding to the slow modulation of the wavetrains. In other words, we allow gradual changes in ω, κ , and a, assuming them to be noticeable only at distances much larger than κ−1 and after times much longer than ω −1 . This is usually known as the geometrical optics approximation. Even though the wave parameters evolve, the general functional form of u is retained in this approach. We consider the (1 + 1)-dimensional, nonlinear Klein–Gordon equation utt − uxx + V (u) = 0,
(11.5.1)
where V (u) is a general nonlinear function of u, but not its derivatives, and the prime denotes the derivative with respect to u. We now multiply (11.5.1) by ut and ux and reorganize the terms to derive the following conservation equations:
1 2 ut + u2x + V + [−ut ux ]x = 0, (11.5.2) 2
1 2 ut + u2x − V = 0. (11.5.3) [−ut ux ]t + 2 x
586
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
We look for a stationary wave solution of (11.5.1) in the form u = u(ξ),
ξ = x − ct
(11.5.4ab)
so that integration of the resulting equation gives
1 2 c − 1 u2ξ + V (u) = A, 2
(11.5.5)
where A is an integration constant related to the amplitude of the waves. So, it follows from equation (11.5.5) that (11.5.1) admits stationary wave solutions of the form u2ξ = f (u, c, A). (11.5.6) Based on the phase plane analysis with the phase diagrams of uξ (ξ), it can be shown that periodic wave solutions exist for all V (u) with a minimum. So we assume that (11.5.6) admits a solution of the form u = u(ξ, c, A) so that (11.5.6) gives
du dξ
2 = A − V (u) 2 (c − 1)
12 ,
in which we assumed that c2 > 1 and A > V (u). Or equivalently,
− 1 1 2 ξ= A − V (u) 2 du. c −1 2
(11.5.7)
(11.5.8)
Clearly, the solution is oscillatory between two consecutive zeros, say u1 (c, A) and u2 (c, A) with u2 > u1 of the function f (u, c, A) between which it is positive definite. The positive definiteness between u1 and u2 is a necessary condition to ensure that ( du dξ ) is real. We suppose that ξ1 and ξ2 are the values of ξ at which u(ξ1 , c, A) = u1 (c, A) and u(ξ2 , c, A) = u2 (c, A). Using the analogy of the linear waves, we define the wavelength λ for nonlinear waves by ξ2 u2
u2 du 1 2 du . λ = λ(c, A) = 2 dξ = 2 = c −1 2 A − V (u) ξ1 u 1 uξ u1 (11.5.9) Similarly, the wavenumber k = k(c, A), the frequency ω = ω(c, A), and the periodic time T = T (c, A) are defined by k=
2π , λ(c, A)
ω = ω(c, A) = ck(c, A),
T = T (c, A) =
2π . ω(c, A) (11.5.10)
In direct analogy with the theory of linear waves, it is possible to find a general solution by considering c and A as slowly varying functions of x and t.
11.5 The Nonlinear Klein–Gordon Equation and Averaging Techniques
587
We now describe the Whitham averaging method to eliminate the rapid fluctuations in the field functions taking place at the smaller space scale x ∼ λ and time scale t ∼ T . The significant changes of order O(1) in k, ω, and a take place over the length and time scales of the order of L = O(x) and T = O(t), respectively. The above discussion indicates the function of intermediate scales X and τ , so that λ X L and T τ T . We now define the local average f (x, t) of f (x, t) at any point x for a fixed t by the relation f (x, t) =
1 2X
x+X
f (x , t) dx .
(11.5.11)
x−X
To develop the averaging method, Whitham used the conservation form of the dynamical equation as ∂ρ + ∇ · qi = 0, ∂t
i = 1, 2, . . . , n.
(11.5.12)
Obviously,
∂ρ ∂t ∂q ∂x
=
1 2X
x+X
x−X x+X
∂ρ ∂ ρ(x , t) dx = , ∂t ∂t
(11.5.13)
1 ∂ 1 q(x , t) dx = q(x + X, t) − q(x − X, t) 2X x−X ∂x 2X x+X 1 ∂ ∂q . (11.5.14) q(x , t) dx = = ∂x 2X x−X ∂x
=
Taking the average of (11.5.12) gives ∂ ρ(x, t, c, A) + ∇ · q i (x, t, c, A) = 0. ∂x
(11.5.15)
The average quantities ρ and q i in (11.5.15) still depend explicitly on x and t. To eliminate this explicit dependence, we consider the fact that, since X λ, the interval (x − X, x + X) contains a large number of waves over which k, ω, etc. are nearly constant. In this case, we can replace ρ and q i by ρ(c, A) and qi (c, A), respectively, which are the averages over one wavelength by keeping k, ω, etc. constant on (x − X, x + X). When this approximation is made, the errors introduced λ are of the order of X ( 1) and X L ( 1), which are very small. We now define the global average f (c, A) of f (x, t) over a wavelength by 1 f (c, A) = λ =
1 λ
x+λ
f (x , t) dx =
x x+λ
f (u, c, A) dX, x
where the integrand depends on u.
1 λ
λ
f (X, t) dX,
x = x + X,
0
(11.5.16)
588
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
In view of the average defined by (11.5.16), the averaged conservation equation is obtained from (11.5.12) as ∂ ρ(c, A) + ∇ · qi (c, A) = 0, ∂t
(11.5.17)
where c and A are slowly varying functions of x and t. It is noted that the terms λ (1) upon differentiation. The average in (11.5.17) become of the order of L
f (c, A) defined by (11.5.16) plays an important role in the study of nonlinear waves. In direct analogy with the adiabatic invariant I = p dq in the Hamiltonian mechanics, it is expedient to introduce a new function W (c, A) = uξ du, (11.5.18) which enables us to express the wavenumber, frequency, and other parameters involved in the wave in terms of the first-order partial derivatives of W with respect to c and A. Furthermore, in many situations, the system of equations involving these partial derivatives of W turns out to be hyperbolic. Consequently, the characteristics of this hyperbolic system define the characteristic speeds associated with the wave. These remarkable features will be discussed further in the context of the Klein– Gordon equation. In view of the result (11.5.7) for the Klein–Gordon equation, it is convenient to introduce the function W (c, A) given by
1 uξ · du = 2 c2 − 1 A − V (u) 2 du 2πW (c, A) = c2 − 1
c2 − 1 G(A), (11.5.19) = where G(A) is another new function defined by
1 A − V (u) 2 du, G(A) = which is independent of c, but depends only on A. Obviously, du 1 λ 1 G (A) = =√ , dξ = √ 2 2 2 c −1 c −1 A − V (u)
(11.5.20)
(11.5.21)
in which (11.5.9) is used, and du du 1 1 1 √ dξ = − G (A) = − √ 3/2 3/2 dξ {A − V (u)} {A − V (u)} 2 2 2 2 λ 1 du < 0, (11.5.22) = 2 (c − 1)3/2 0 dξ in which (11.5.7) is again utilized.
11.5 The Nonlinear Klein–Gordon Equation and Averaging Techniques
Further, it follows from (11.5.19) and (11.5.21) that
∂W 2π = . c2 − 1 G (A) = λ = 2π ∂A k This shows that the wavenumber k can be defined by the relation ∂W = 1. k ∂A
589
(11.5.23)
(11.5.24)
We use (11.5.16) to calculate the average values of all quantities involved in equations (11.5.2) and (11.5.3) as follows:
2
1 2 1 2 1 1 2 2 ut + ux c + 1 uξ = c2 + 1 u dξ = 2 2 2 λ ξ
1 2 1 c2 + 1 kW (c, A), (11.5.25) c +1 = kuξ du = 4π 2 c2 − 1 in which (11.5.19) is used. Furthermore, we find that c kW,
−ut ux = cu2ξ = 2 (c + 1)
2 1 2 1 V (u) = A − c + 1 uξ = A − kW. 2 2
(11.5.26) (11.5.27)
Differentiating (11.5.19) partially with respect to c gives c ∂W = 2 W. ∂c (c + 1)
(11.5.28)
We substitute the above average values in (11.5.2) and (11.5.3) and then utilize results (11.5.24) and (11.5.28) to derive the following averaged conservation equation: ∂ ∂ k(cWc + AWA − W ) + kc(cWc + AWA − W ) − cA = 0, (11.5.29) ∂t ∂x ∂ ∂ [kWc ] + [ckWc − A] = 0, (11.5.30) ∂t ∂x where, in writing the second term in (11.5.29), we have added kc AWA and subtracted its equivalent expression cA. We next calculate partial derivatives of various terms in (11.5.29) and collect the coefficients of c, A, and W to obtain ∂ ∂ ∂ ∂ (kWc ) + (ckWc − A) + A (kWA ) + (ckWA − c) c ∂t ∂x ∂t ∂x ∂ ∂k + (ck) = 0. (11.5.31) −W ∂t ∂x
590
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
The first term in this equation vanishes due to (11.5.30), the second term also vanishes since kWA = 1, and hence, finally, we obtain ∂ ∂k + (ck) = 0, ∂t ∂x
(11.5.32)
which is, introducing the frequency ω = ck, ∂k ∂ω + = 0. ∂t ∂x
(11.5.33)
Or equivalently, ∂k ∂k + ω (k) = 0. (11.5.34) ∂t ∂x This is the fundamental kinematic equation of the wave. Thus, along the characteris tics dx dt = ω (k), the wavenumber k (and hence, frequency ω) is conserved. Substitution of k = WA−1 in (11.5.32) yields ∂WA ∂c ∂WA +c − WA = 0. ∂t ∂x ∂x
(11.5.35)
Similarly, expanding (11.5.30) combined with (11.5.24) leads to the result ∂Wc ∂A ∂Wc +c − WA = 0. ∂t ∂x ∂x
(11.5.36)
Finally, the two independent averaged conservation equations (11.5.35), (11.5.36) can be expressed in the form DWA ∂c − WA = 0, Dt ∂x DWc ∂A − WA = 0, Dt ∂x
(11.5.37) (11.5.38)
where the total derivative is defined by D ∂ ∂ = +c . Dt ∂t ∂x
(11.5.39)
To determine the characteristics and characteristic speeds, we use (11.5.32) and (11.5.28) and, then, replace W in terms of G to express (11.5.37), (11.5.38) in terms of c and A as cG cG ct + 2 cx = 0, − 1) (c − 1) G cG cG At + G Ax − 2 ct − 2 cx = 0. (c − 1) (c − 1)
G At + cG Ax +
(c2
These equations admit the following two characteristics C± , so that
(11.5.40) (11.5.41)
11.5 The Nonlinear Klein–Gordon Equation and Averaging Techniques
C+ :
dx = dt
1 + cα c+α
and C− :
where
α=
−
dx = dt
GG G2
1 − cα , c−α
591
(11.5.42ab)
12 (11.5.43)
.
The associated compatibility relations are given by 1 dc G 2 − − dA = 0 (c2 − 1) G 1 G 2 dc + − dA = 0 (c2 − 1) G
along C+ ,
(11.5.44)
along C− .
(11.5.45)
These relations define two Riemann invariants 1 c A dc G 2 − r= − dA, 2 G c0 (c − 1) A0 1 A c dc G 2 − + dA. s= 2 G c0 (c − 1) A0
(11.5.46) (11.5.47)
Equations (11.5.42ab) reveal that there are two characteristic speeds (1 ± cα)/ (c±α). If we consider a wavetrain which is initially uniform with c = c0 and A = A0 outside some finite region, then, after some interaction period, the disturbance splits into two simple waves separated by a domain of constant values of c and A. In one simple wave, the characteristics C+ are straight lines which carry constant values of r, and the other Riemann invariant is constant everywhere. On the other hand, in the second simple wave, r has the same value everywhere, and s is constant along the characteristics C− , which are straight lines. Between these two simple waves, c and A assume constant values. Since the wavenumber and amplitude are expressible in terms of c and A, the qualitative features similar to those as stated above are also applicable to them. The energy propagation speed is then obtained from (11.5.29) as the ratio of energy flux to energy density and has the form kc(cWc + AWA − W ) − cA k(cWc + AWA − W )
kc(cWc − W ) + cA − cA by (11.5.24) = k(cWc − W ) + A
cW c(cWc − W ) = = by (11.5.28) , A A 2 (cWc − W ) + k W + (c − 1)( k ) which, by results (11.5.19) and (11.5.21), is =
G+
cG . − 1)AG
(c2
(11.5.48)
592
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
This is another important new speed which is definitely not one of the two characteristic speeds. This new speed does not seem to have a simple physical interpretation. Thus, in the present fully nonlinear case, there exist three important speeds— two characteristic speeds and one energy propagation speed. Clearly, one of them should be interpreted as the group velocity.
11.6 The Klein–Gordon Equation and the Whitham Averaged Variational Principle The Lagrangian density of the Klein–Gordon equation (11.5.1) is given by L=
1 2 ut − u2x − V (u). 2
(11.6.1)
We seek a progressive, periodic wave solution of the nonlinear Klein–Gordon equation (11.5.1) in the form u = u(θ),
θ = kx − ωt,
(11.6.2)
where the local wavenumber and local frequency are given by k = θx and ω = −θt , respectively. Substituting (11.6.2) in equation (11.5.1) gives
2 (11.6.3) ω − k2 uθθ + V (u) = 0, which gives the integral
1 2 ω − k 2 u2θ + V (u) = A, 2
(11.6.4)
where A is a constant of integration. This equation can be solved by a quadrature for a suitable form of V (u) and has a solution, periodic in θ, given by θ=
1 2 ω − k2 2
12
du A − V (u)
.
(11.6.5)
Since u is a periodic function with period 2π, we require that 2π =
1 2 ω − k2 2
12
du , A − V (u)
(11.6.6)
where denotes integration over a complete period. Result (11.6.6) represents a relation between ω, k, and A, and hence, it represents the dispersion relation. According to the Whitham theory, the wave parameters k, ω, and A are slowly varying functions of x and t corresponding to the slow modulation of the wave. These parameters can then be determined from the averaged Lagrangian defined by Whitham (1965a, 1965b) as
11.6 The Klein–Gordon Equation and the Whitham Averaged Variational Principle
L=
1 2π
593
2π
(11.6.7)
L dθ, 0
where L is given by (11.6.1) which is simplified by using (11.6.2) so that L=
1 2 ω − k 2 u2θ − V (u) = ω 2 − k2 u2θ − A. 2
(11.6.8)
Substituting (11.6.8) in (11.6.7) gives
2π ∂u 2 1 2 A 2π 2 L(ω, k, A) = dθ − dθ ω −k 2π ∂θ 2π 0 0
2π ∂u 1 2 = du − A, ω − k2 2π ∂θ 0 which is, by (11.6.4),
1 1 2 ω − k2 2 A − V (u) du − A 2π
1 1 2 ω − k2 2 G(A) − A, = 2π
=
(11.6.9)
where G(A) is defined by (11.5.20). According to the Whitham theory, wave parameters ω, k, and A are determined from the averaged variational principle δ L(ω, k, A) dx dt = 0, (11.6.10) where ω and k are related by ω = −θt and k = θx , and hence, they cannot be varied independently. The variational equations from variations δA and δθ in (11.6.10) are given by ∂L = 0, ∂A ∂ ∂L ∂ ∂L ∂ ∂ ∂L δθ : + = − (Lω ) + = (Lk ) = 0. ∂θ ∂t ∂θt ∂x ∂θx ∂t ∂x
δA :
(11.6.11) (11.6.12)
Thus, by (11.6.9), equation (11.6.11) becomes
1 1 2 ω − k2 2 G (A) = 1, 2π
(11.6.13)
or equivalently, k=
2π (c2
−
1)G (A)
,
2πc ω = ck = . 2 (c − 1)G (A)
In terms of G(A), we can express kt + ωx = 0 in the form
(11.6.14ab)
594
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
∂ ∂ 1 c √ √ + = 0. ∂t ∂x c2 − 1G (A) c2 − 1G (A) Similarly, the amplitude equation (11.6.12) can be expressed as ∂ cG(A) ∂ G(A) √ √ + = 0. ∂t ∂x c2 − 1 c2 − 1
(11.6.15)
(11.6.16)
The Whitham Lagrangian formulation is relatively more simple than the averaging procedure discussed in the previous section. Indeed, the averaged Lagrangian is a useful quantity in nonlinear theory whenever it is available. It is also important to point out that the present analysis is applied to slowly varying nonlinear waves. However, the variational method is applicable to weak nonlinear waves and also to all nonlinear wavetrains. In particular, for the linear Klein–Gordon equation, V (u) = u so that V (u) = 1 2 2 u . Hence the function G(A) defined by (11.5.20) can easily be computed by using the linear solution for u = −a cos θ where 2A = a2 . It turns out that G(A) = 2πA so that G (A) = 2π. Consequently, the averaged Lagrangian L can be obtained explicitly from (11.6.9) as L=
ω2 − k2
12
− 1 A,
A=
1 2 a , 2
(11.6.17)
so that the linear dispersion relation is LA = 0, that is, ω2 − k 2 = 1.
(11.6.18)
This is independent of the amplitude A, as expected in the linear theory. This also confirms the fact that the Whitham variational method is applicable to both linear and nonlinear wave propagation problems.
11.7 The Sine-Gordon Equation: Soliton and Antisoliton Solutions The sine-Gordon equation has a long history that begins in the latter part of the nineteenth century when this equation was discovered in differential geometry in connection with the theory of surfaces of constant negative curvature. Various methods for finding particular solutions of this equation were developed at that time. One of the methods is known as the Bäcklund transformations, which will be considered in a subsequent section. Other methods include traveling wave solutions, the similarity method, the inverse scattering method, and the method of separation of variables, which deals with the representation of solutions as functions of independent variables. All of these methods will be discussed in subsequent sections. It has already been indicated in Chapter 2 that the sine-Gordon equation is one of the basic nonlinear evolution equations and has recently been employed to describe
11.7 The Sine-Gordon Equation: Soliton and Antisoliton Solutions
595
various important nonlinear physical phenomena. The standard form of this equation is 1 ∂2u ∂ 2u − = sin u. (11.7.1) ∂X 2 c2 ∂T 2 In terms of the characteristic coordinate transformations 1 x = ± (X − cT ), 2
t=
1 (X + cT ), 2
(11.7.2ab)
equation (11.7.1) takes the following form: uxt = sin u.
(11.7.3)
Thus, the two forms (11.7.1) and (11.7.3) of the sine-Gordon equation are interchangeable. The equation is invariant under u → u + 2nπ, where n = ±1, ±2, ±3, . . . . The transformation u → u + (2n + 1)π leads to the replacement of sin u by − sin u in the equation. The linearized form of equation (11.7.1) is given by uXX − c−2 uT T = u.
(11.7.4)
A plane wave solution of the form u(X, T ) = A exp[i(kX − ωT )] exists provided the dispersion relation
ω 2 = c2 1 + k2 (11.7.5) is satisfied. This shows that ω is real for all real k, and the equilibrium solution u = 0 of (11.7.1) is stable. On the other hand, if u = u ˜ + π for a small perturbation u ˜, (11.7.1) can be linearized to obtain ˜T T = −˜ u. u ˜XX − c−2 u
(11.7.6)
A plane wave solution of the form u ˜(X, T ) = A exp[i(kX − ωT )] also exists, provided the dispersion relation
(11.7.7) ω2 = c2 k 2 − 1 ˜(X, T ) will grow exponentially in is satisfied. Hence ω2 < 0, if 0 ≤ k 2 < 1, and u time T , that is, the equilibrium solution u = π is definitely unstable. Physically, this is quite natural because, when the term uXX is absent (that is, when k = 0), the sineGordon equation (11.7.1) reduces to the equation of a simple pendulum representing finite oscillations. Thus, the question of stability or instability is an important feature of the equilibrium solutions of the sine-Gordon equation. We next discuss the single-soliton solutions of the sine-Gordon equation 1 ∂2u ∂2u − = sin u. ∂x2 c2 ∂t2
(11.7.8)
596
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
We seek a solution of the form u(x, t) = φ(x − U t) = φ(ξ) which corresponds to a wave traveling with a velocity U and, then, substitute in (11.7.8) to obtain an ordinary differential equation for φ(ξ) in the form 2 d2 φ 2d u − V = sin φ, dξ 2 dξ 2
where V 2 =
U2 . c2
(11.7.9)
Dividing both sides by (1 − V 2 ) and multiplying by φ (ξ) gives sin u φ (ξ)φ (ξ) = φ (ξ). (11.7.10) 1−V2
Or equivalently,
that is,
2 cos u d 1 dφ = 0, + dξ 2 dξ 1−V2
(11.7.11)
2 cos u 1 dφ + = B = const. 2 dξ (1 − V 2 )
(11.7.12)
Solving for φ (ξ) gives a first-order, ordinary differential equation for φ(ξ) 1 2 cos φ 2 2A − . 1−V2
dφ = dξ
(11.7.13)
We can separate the variables and, then, integrate to obtain
φ
φ0
√
dψ = A − cos ψ
2 1−V2
12
ξ
dη,
(11.7.14)
ξ0
√ where A = B 1 − V 2 . This result depends on the two parameters V (or U = cV ), the velocity of the soliton, and B, an integrating constant. Thus, the stable solutions represent solitary waves, periodic waves, or a monotonically increasing function of ξ (Scott 1969) depending on the sign and magnitudes of U and A. When A = 1, a solitary wave solution exists for any velocity, so that 0 < |V | < 1 (or 0 < |U | < c). In this case, we use the trigonometric identities 1 − cos φ = 2 sin2 ( 12 φ) d and dφ [log tan( 14 φ)] = (2 sin 12 φ)−1 to simplify both sides of (11.7.14), so that
2 1−V2
12
√ (ξ − ξ0 ) = 2
φ
φ0
√ tan( 14 φ) dψ . = 2 log 2 sin 12 ψ tan( 14 φ0 )
(11.7.15)
This leads to the solution for φ(ξ) as −1
φ(ξ) = 4 tan
ξ − ξ0 α exp √ 1−V2
,
(11.7.16)
11.7 The Sine-Gordon Equation: Soliton and Antisoliton Solutions
597
Fig. 11.1 The soliton and antisoliton solutions.
where α = tan( 14 φ0 ) and ξ0 are constants of integration which can be incorporated in several ways. We choose α = 1 and ξ0 = 0 to obtain one simple solitary wave solution for u(x, t) in the form x−Vt −1 exp √ . (11.7.17) u(x, t) = 4 tan 1−V2 This is called the soliton (or kink) solution of the sine-Gordon equation and represents a continuous profile with u → 0 as x → −∞, and u → 2π as x → +∞, shown in Figure 11.1. The soliton propagates in the positive x-direction with velocity V . Another solution can be obtained from (11.7.15) in the form x−Vt x−Vt −1 −1 exp − √ = 4 cot exp √ . u(x, t) = 4 tan 1−V2 1−V2 (11.7.18) This is called the antisoliton (or antikink). This solution profile travels in the negative x-direction with velocity V . It represents a continuous profile with u → 2π as x → −∞, and u → 0 as x → +∞, also shown in Figure 11.1. The partial derivatives of u(x, t), that is, ux and ut , also represent solitary waves which are given by ux (x, t) = φ (x − U t)
and
ut (x, t) = −U φ (x − U t),
(11.7.19ab)
where φ (ξ) is obtained from (11.7.16) as √ 2sech{(x − V t)/ 1 − V 2 } √ = φ (ξ) = √ 1 − V 2 {1 + tan2 ( 14 φ)} 1−V2
4 tan( 14 φ)
t in which tan( 14 φ) is replaced by exp( √x−V ). 1−V 2 If A = 1, other types of solutions of the sine-Gordon equation can be obtained. In particular, when V > 1 and |A| < 1, the integral in (11.7.14) with φ0 = 0 and ξ0 = 0 can be written as
598
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
ξ= =
1−V2 2 1−V 2
12
φ
dψ (1 − cos ψ) − (1 − A)
φ
dψ , √ 2ψ 2 sin 2 − m2
0
1 2 2
0
sin( ψ2 ), is 2
12 = V −1
which, substituting s =
m2 =
1 (1 − A), 2
1 m
0
s
√
1−
s2
ds √ . 1 − s2 m2
(11.7.20)
This can be expressed in terms of Jacobi’s elliptic function (Dutta and Debnath 1965) as φ ξ 1 sin = sn √ ,m , (11.7.21) s= m 2 V2−1 where m is the modulus of Jacobi’s elliptic function sn(z, m). Thus, the final solution is given by ξ ,m . (11.7.22) u(x, r) = φ(ξ) = 2 sin−1 msn √ V2−1
11.8 The Solution of the Sine-Gordon Equation by Separation of Variables The form of the solution (11.7.17) suggests that the transformation 1 v(x, t) = tan u 4
(11.8.1)
can be used to transform the sine-Gordon equation uxx − utt = sin u
(11.8.2)
1 + v 2 (vxx − vtt − v) − 2v vx2 − vt2 − v 2 = 0,
(11.8.3)
into the form
in which the trigonometric identity sin u = 4v(1 − v2 )/(1 + v 2 )2 has been utilized. We seek solutions by the separation of variables in the form 1 φ(x) v(x, t) = tan u = , (11.8.4) 4 ψ(t) for some functions φ(x) and ψ(t) to be determined. Substituting (11.8.4) in equation (11.8.3) and using the identity sin 4θ = 4 tan θ(1 − tan2 θ)/(1 + tan2 θ)2 , we obtain
11.8 The Solution of the Sine-Gordon Equation by Separation of Variables
φ2 + ψ 2
φxx
ψtt − 2 φ2x + ψt2 = φ2 − ψ 2 . + φ ψ
599
(11.8.5)
Differentiating this equation with respect to x and t enables us to separate the variables so that 1 1 φxx ψtt =− = −4κ2 , (11.8.6ab) (φφx ) φ x (ψψt ) ψ t where −4κ2 is a separation constant. Each of these ordinary differential equations (11.8.6ab) can be integrated twice to find φ2 = −κ2 φ4 + aφ2 + b
and ψ 2 = κ2 ψ 4 + cψ 2 + d.
(11.8.7ab)
Substituting these equations in (11.8.5) reveals that a − c = 1 and b + d = 0. Setting a = m2 and b = n2 , we obtain
φ2x = −κ2 φ4 + m2 φ2 + n2 and ψt2 = κ2 ψ 4 + m2 − 1 ψ 2 − n2 , (11.8.8ab) where m and n are integration constants. In general, these equations can be solved in terms of elliptic functions. However, we solve these equations for the following special cases of interest: Case 1. κ = n = 0 and m > 1. In this case, equations (11.8.8ab) take the form φx = ±mφ and ψt = ± m2 − 1ψ, which give exponential solutions and
φ(x) = a1 exp(±mx)
ψ(x) = a2 exp ± m2 − 1 t ,
(11.8.9ab)
(11.8.10ab)
where a1 and a2 are integration constants. Thus, the solution (11.8.4) becomes x ± Ut −1 u(x, t) = 4 tan α exp ± √ , (11.8.11) 1 − U2 √
where α = aa12 and U = mm−1 (or m = (1 − U 2 )− 2 ) are constants. Evidently, one of these solutions (11.8.11)
u(x, t) = 4 tan−1 α exp m(x − U t) 2
1
(11.8.12)
is identical with (11.7.17) for α = 1, and hence, it represents the soliton (or kink) solution of the sine-Gordon equation. On the other hand, from (11.8.11), another solution can be found in the form
−1 −1 1 u(x, t) = 4 tan α exp −m(x − U t) = 4 cot exp m(x − U t) . α (11.8.13)
600
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
This is also identical with (11.7.18) for α = 1 and represents the antisoliton (or antikink) solution of (11.8.2). Furthermore, ux and ut also represent solitary wave solutions given by (11.8.14) ux (x, t) = ±2m sech m(x ± U t) + log α
and ut (x, t) = ±2
m2 − 1 sech m(x ± U t) + log α .
(11.8.15)
Two other choices in sign in (11.8.11) represent antisoliton (or antikink) solutions of the sine-Gordon equation. Case 2. κ = 0, m2 > 1, and n = 0. In this case, solutions of (11.8.8ab) are obtained by integration in terms of hyperbolic functions as n φ(x) = ± sech(mx + a1 ), m (11.8.16)
n 2 ψ(t) = √ cosh m − 1 t + a2 , m2 − 1 where a1 and a2 are integrating constants. Thus, the solution is given by U sinh(mx + a1 ) √ u(x, t) = ± tan−1 . cosh( m2 − 1t + a2 )
(11.8.17)
Since this result is made up of the ratio (φ/ψ), it does not depend on n. In particular, when a1 = a2 = 0, solution (11.8.17) can be expressed in the form −1 U sinh(mx) , 0 < U 2 < 1. u(x, t) = ±4 tan (11.8.18) cosh(mU t) This solution was first discovered by Perring and Skyrme (1962) based on numerical calculations for two interacting solitons. They verified analytically that (11.8.18) is, indeed, the exact solution, and it describes the interaction of two solitons similar to those of the KdV solitons, where solitons emerge from the interaction only with a slight change of phase. To show that the solution describes the interaction of two equal solitons, we investigate it asymptotically as t → ±∞. Expressing sinh(mx) in terms of a sum of two exponentials, we first note that cosh(mU t) ∼
1 exp mU |t| 2
as t → ±∞.
In view of these results, it follows at once that
v(x, t) ∼ U exp m(x + U t) − U exp −m(x − U t)
as t → −∞. (11.8.19)
11.8 The Solution of the Sine-Gordon Equation by Separation of Variables
601
Fig. 11.2 Interaction of two equal solitons (11.8.18) for U > 0. (a) t → −∞, (b) t = 0, and (c) t → +∞. From Drazin (1983).
Solution (11.8.19) is uniformly valid for all x and represents two distinct solitons approaching one another at equal but opposite speed U , as shown in Figure 11.2. More precisely, in the limit as x → −∞ and t → −∞, the solution becomes
u(x, t) ∼ −4 tan−1 U exp −m(x − U t) . (11.8.20) Solution (11.8.20) represents a pulse moving in the positive x-direction, as u(x, t) increases from −2π to 0, as x passes through the value U t. As x → +∞ and t → −∞, the solution reduces to the form
(11.8.21) u(x, t) ∼ 4 tan−1 U exp −m(x + U t) . This also represents a pulse traveling in the negative x-direction, as u(x, t) increases from 0 to 2π, as x passes through the value −U t. At t = 0, two pulses suffer from an interaction. All of these results are shown in Figure 11.2. Similarly, as t → ∞, the asymptotic solution is given by x − Ut x + Ut + U exp − √ . (11.8.22) v(x, t) ∼ −U exp − √ 1 − U2 1 − U2 In other words, as x → −∞ and t → +∞, this asymptotic solution is x + Ut −1 u(x, t) ∼ −4 tan U exp − √ . (11.8.23) 1 − U2
602
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
As x → +∞ and t → +∞, this asymptotic solution for u(x, t) is given by x − Ut −1 U exp √ . (11.8.24) u(x, t) ∼ 4 tan 1 − U2 Since u varies from −2π to +2π as x changes from −∞ to +∞, the corresponding solution is called a 4π pulse. Finally, the only visible evidence of the interaction remaining, as t → +∞, is a longitudinal displacement of each soliton. To show this, we observe the exponential form of the solution 1 (11.8.25) ±U exp ±m(x + U t) = ± exp ±m x + U t ± γ , 2 √ where γ = 2 1 − U 2 log(U −1 ) represents the displacement of each soliton which is retarded by the interaction. The above soliton solutions produce ux and ut . Clearly, for U > 0, we find ux (x, t) = √
4U cosh(mx) cosh(mU t) . 1 − U 2 [sinh2 (mx) + cosh2 (mU t)]
(11.8.26)
This represents the interaction of two hump-shaped solitons as shown in Figure 11.3. This also shows, more clearly, the interaction of two solitons for t → ±∞. Case 3. κ = 0, n = 0, and m2 > 1. In this case, the solution u(x, t) is given by u(x, t) = −4 tan
−1
√ m sinh( m2 − 1 t + a2 ) √ . · cosh(mx + a1 ) m2 − 1
(11.8.27)
This result is somewhat similar to that of (11.8.17), and can be interpreted physically with a given boundary condition ux (0, t) = 0 for all time t. It represents a soliton moving toward the boundary and reflected back as an antisoliton, and hence, solution (11.8.27) represents an interaction of a soliton with an antisoliton. In the limit as m → 1, solution (11.8.27) reduces to the form u(x, t) ∼ −4 tan−1 (t + a2 )sech (x + a1 ) . (11.8.28) Case 4. κ = 0, n = 0, and m2 < 1. This case gives a new solution which can be obtained from sinh(ωt + a2 ) m −1 √ , (11.8.29) · u(x, t) ∼ −4 tan 1 − m2 cosh(mx + a1 ) √ where ω = 1 − m2 = mU . This is known as the breather solution of the sineGordon equation and represents a pulse-type structure of a soliton. For fixed x, the
11.8 The Solution of the Sine-Gordon Equation by Separation of Variables
603
Fig. 11.3 Interaction of two hump-shaped solitons for (a) t → −∞, (b) t = 0, and (c) t → √ +∞ with amplitude A = 2U/ 1 − U 2 . From Drazin (1983).
√ solution is a periodic function of time t with frequency ω = 1 − m2 . For m = 0.8, the breather solution as a function of x is shown in Figure 11.4. The case for m 1 corresponds to a small-amplitude breather solution. This can be derived from (11.8.29) by expanding the inverse tangent function for small m and retaining only the first term. This leads to a limiting solution in the form m2 t + c.c., (11.8.30) u(x, t) ∼ 2im sech(mx) exp i 1 − 2 = A(x, t) exp(it), (11.8.31) 2
where A(x, t) = 2im sech(mx) exp( itm 2 ). It can easily be verified that A(x, t) satisfies the (1 + 1)-dimensional nonlinear Schrödinger equation 2iAt − Axx − |A|2 A = 0,
(11.8.32)
where the term Att = O(m4 ) has been neglected. Finally, more general solutions of equations (11.8.8ab) can be found in terms of elliptic functions. It is also of interest to investigate the solutions of the sine-Gordon equation confined to a finite region of space by boundaries. It turns out that the oscillations of the soliton’s position closely resemble those of a particle confined by
604
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
Fig. 11.4 Breather solution of (11.8.2) for m = 0.8.
an external potential. This confirms the fact that solitons behave like elastic particles. For details, the reader is referred to De Leonardis and Trullinger (1980) and Newell and Kaup (1978). It may be noted that a variety of solutions of the sine-Gordon equation have been studied by Lamb (1973, 1980). Example 11.8.1 (Energy of a Soliton, an Antisoliton, and a Breather). We consider the energy associated with the soliton and the antisoliton of the sine-Gordon equation uxx − utt = sin u.
(11.8.33)
The Lagrangian density of this equation is given by L=
1 2 ut − u2x − (1 − cos u). 2
(11.8.34)
The Hamiltonian density has the form H = ut
∂L ∂ut
− L,
(11.8.35)
which, due to (11.8.34), is =
1 2 ut + u2x + (1 − cos u). 2
(11.8.36)
For the single-soliton solution
u(x, t) = 4 tan−1 exp m(x − U t) ,
(11.8.37)
where m = (1 − U 2 )− 2 , we obtain the Hamiltonian density H and energy E as
(11.8.38) H = 4m2 sech2 m(x − U t) , 1
11.9 Bäcklund Transformations for the Sine-Gordon Equation
∞
E= −∞
H dx = 8m.
605
(11.8.39)
For the antisoliton solution
u(x, t) = 4 tan−1 exp −m(x − U t) ,
(11.8.40)
the energy E is also equal to 8m. This shows that the solution represents the elastic interaction of a soliton and an antisoliton, each of which has energy 8m. For the soliton–antisoliton solution (11.8.27), we find that 2 sech2 {mx(p + q tanh2 mx)} 8m , (11.8.41) · H= κ (1 + rsech2 mx)2 where κ2 = m2 − 1 > 0, p = κ2 + (1 + κ2 )sech2 (κt), q = m2 sinh2 (κt), and 2 2 r = (m κ ) sinh (κt). In this case, it is easy to check that the energy E = 16m, where m > 1. Finally, the breather solution is given by (11.8.29). The associated energy is E = 16m, where m > 1. The energy is, thus, equal to the sum of energies of two free solitons (or antisolitons) and leads to the physical interpretation of the breather as a bound state composed of a pair of a soliton and an antisoliton.
11.9 Bäcklund Transformations for the Sine-Gordon Equation We consider the characteristic form of the sine-Gordon equation uxt = sin u.
(11.9.1)
In 1880, Bäcklund discovered the transformation as the pair of equations 1 1 (u − v)x = a sin (u + v), 2 2 1 1 1 (u + v)t = sin (u − v), 2 a 2
(11.9.2a) (11.9.2b)
where a is a nonzero arbitrary real constant. This pair of equations is called the Bäcklund transformations of (11.9.1). The cross-differentiation of (11.9.2a), (11.9.2b) gives a 1 1 1 1 (u − v)tx = (u + v)t cos (u + v) = sin (u − v) cos (u + v), 2 2 2 2 2 1 1 1 1 1 (u + v)xt = (u − v)x cos (u − v) = sin (u + v) cos (u − v). 2 2a 2 2 2
(11.9.3) (11.9.4)
Adding and subtracting these equations, we obtain uxt = sin u
and
vxt = sin v.
(11.9.5ab)
606
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
Evidently, the Bäcklund transformations (11.9.2a), (11.9.2b) relate two solutions u and v that satisfy the same equation (11.9.1). Since both u and v satisfy the same sine-Gordon equation, the pair (11.9.2a), (11.9.2b) is referred to as an auto-Bäcklund transformation for (11.9.1). The auto-Bäcklund transformation can be used to construct a sequence of solutions of the sine-Gordon equation beginning with any given solution. We next use (11.9.2a), (11.9.2b) to solve the sine-Gordon equation (11.9.1). Obviously, u(x, t) = 0 is a trivial solution of (11.9.1) for all x and t. We use this solution to construct a nontrivial solution. We set v = 0, so that the Bäcklund transformation (11.9.2a), (11.9.2b) becomes 1 2 1 ut = sin u . (11.9.6ab) ux = 2a sin u , 2 a 2 Integrating these equations gives u 1 tan 1 u + A(t), u du = 2 log cosec 2ax = 2 4 u 1 2t 1 = cosec u du = 2 logtan u + B(x), a 2 4
(11.9.7) (11.9.8)
where A(t) and B(x) are arbitrary functions involved as a result of integration. It follows from (11.9.7) and (11.9.8) that u t tan = α exp ax + , (11.9.9) 4 a where α is a constant. Or equivalently,
t u(x, t) = 4 tan−1 α exp ax + . a
(11.9.10)
This represents a new solution of the sine-Gordon equation (11.9.1) and it describes a soliton (or kink) solution. Returning to the original sine-Gordon equation (11.7.1) with x = 12 (X − cT ) and t = 12 (X + cT ), we obtain a 1 (X − cT ) + (X + cT ) u(X, T ) = 4 tan−1 α exp 2 2a
−1 = 4 tan α exp ±m(X − U T ) , (11.9.11) where 12 (a + a1 ) = m = (1 − U 2 )− 2 , |U | < 1, and 2c (a − a1 ) = U m so that 2 U a −1 a=m 1+ and U = c 2 . c a +1 1
Thus the solution (11.9.11) represents a soliton (or kink) and an antisoliton (or antikink) corresponding to the positive or negative sign in (11.9.11). These solutions are of width m, and they propagate with the constant velocity U .
11.9 Bäcklund Transformations for the Sine-Gordon Equation
607
We next construct a new soliton solution representing two interacting solitons by means of a purely algebraic method from a known solution u0 . We first replace u and v by u1 and u2 , respectively, in (11.9.2a), (11.9.2b), and, then, obtain two solutions corresponding to two arbitrary constants a = a1 and a = a2 with a1 = a2 . Using these solutions, we seek two more solutions u3 and u4 corresponding to parameters a = a2 and u2 with a = a1 and u2 , respectively. According to the above procedure, we obtain four relations from (11.9.2a), (11.9.2b): ∂u1 ∂x ∂u2 ∂x ∂u3 ∂x ∂u4 ∂x
∂u0 ∂x ∂u0 = ∂x ∂u1 = ∂x ∂u2 = ∂x =
1 + 2a1 sin (u1 + u0 ), 2 1 + 2a2 sin (u2 + u0 ), 2 1 + 2a2 sin (u3 + u1 ), 2 1 + 2a1 sin (u4 + u2 ). 2
(11.9.12) (11.9.13) (11.9.14) (11.9.15)
We set u3 = u4 = v2 in (11.9.14) and (11.9.15) and choose suitable values for integrating constants. Then, subtracting the resulting expressions (11.9.15) from (11.9.12), and (11.9.14) from (11.9.13), we obtain 1 1 1 1 a1 sin (v2 + u2 ) − sin (u1 + u0 ) = a2 sin (v2 + u1 ) − sin (u2 + u0 ) . 2 2 2 2 This result can be simplified by using the standard formula for the difference of two sine functions to obtain
1 1 (v2 − u0 ) − (u1 − u2 ) = a2 sin (v2 − u0 ) + (u1 − u2 ) . a1 sin 4 4 Further simplification of this result gives a1 + a2 1 1 tan (u1 − u2 ) , tan (v2 − u0 ) = 4 a1 − a 2 4
(11.9.16)
or equivalently, v2 = 4 tan−1
a1 + a 2 a1 − a 2
tan
1 (u1 − u2 ) 4
+ u0 .
(11.9.17)
This gives a new solution in terms of given solutions u0 , u1 , and u2 and it can be regarded as the nonlinear superposition principle for the sine-Gordon equation. It is important to point out that the same solution (11.9.17) can be derived from (11.9.2b). It is also possible to continue this procedure many times, and hence, this leads to an n interacting soliton solution vn from an (n − 1)-soliton solution vn−1 . In particular, the solution for two interacting solitons can be obtained in the form 1 a1 + a2 sinh[ 12 (ξ1 − ξ2 )] tan v2 = , (11.9.18) 4 a1 − a2 cosh[ 12 (ξ1 + ξ2 )]
608
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
where
− 1 ξr = ± 1 + Ur2 2 (X − U T ),
1 ar = ± (1 − Ur )/(1 + Ur ) 2 ,
r = 1, 2,
and solutions (11.9.11) and (11.9.16) have been used. We next set U1 = −U2 = U to reduce the solution (11.9.18) to the form 1 U sinh(mX) tan v2 = . (11.9.19) 4 cosh(mU t) This is exactly the same result (11.8.18) obtained by Perring and Skyrme (1962). It was already investigated as the asymptotic behavior of (11.9.19), as t → ±∞, in Section 11.8. The asymptotic solution describes the interaction of two-soliton solutions.
11.10 The Solution of the Sine-Gordon Equation by the Inverse Scattering Method We use the characteristic form of the sine-Gordon equation (11.9.1). Following Lamb (1980), we describe the method of solution of (11.9.1) by the inverse scattering approach. The essence of this approach is to relate equation (11.9.1) to a pair of linear scattering equations for v1 and v2 , ∂v1 1 ∂x + iζv1 = − 2 ux v2 , (11.10.1ab) ∂v2 1 − iζv = u v . 2 x 1 ∂x 2 These equations, together with the conditions that v1 and v2 are bounded as x → ±∞, may be considered as a problem of determining the eigenvalue ζ for a given function q = − 12 ux called the potential corresponding to a solution u(x, t) of (11.9.1). We note that, if v1 and v2 are bounded and q → 0 sufficiently rapidly as x → ±∞, then v1 ∼ a1 exp(−iζx) and v2 ∼ a2 exp(iζx), as x → ±∞. So, there exists an unbounded state if and only if ζ is real. If ζ is taken as a fixed, real number, then ζt = 0. Differentiating (11.10.1ab) with respect to t, integrating with respect to x, and then, combining with equation (11.9.1) gives the evolution equations for the vector eigenfunctions (v1 , v2 ) in the form ∂v1 ∂t
=
i 4ζ (v1
cos u − v2 sin u),
∂v2 ∂t
=
i 4ζ (v1
sin u + v2 cos u).
Since the solution u is related to q through q = − 12 ux so that x u(x, t) = −2 q(ξ, t) dξ, −∞
(11.10.2ab)
(11.10.3)
11.10 The Solution of the Sine-Gordon Equation by the Inverse Scattering Method
609
it follows that u → 0 as x → −∞. Hence, equations (11.10.2ab) give i v1 , 4ζ i ∼ − v2 4ζ
v1t ∼ v2t
(11.10.4a) as x → −∞.
(11.10.4b)
The form of the equations (11.10.2ab) as x → +∞ depends on the value of u(x, t) as x → +∞. The only cases that can be solved without any problem are those in which u(x, t) → 2nπ as x → +∞. it ) as x → −∞. Similarly, From (11.10.4a) v1 is either zero or tends to exp(− 4ζ it ) as x → −∞. We then (11.10.4b) shows that v2 is either zero or tends to exp(− 4ζ consider the solution that is proportional to the fundamental solutions given by x φ1 (x, k) = exp(−ikx) + A1 (x, ξ) exp(−ikξ) dξ (11.10.5) −∞
and φ2 (x, k) =
x
−∞
A2 (x, ξ) exp(−ikξ) dξ,
(11.10.6)
where A1 and A2 represent the scattering (or wake) and the reflected particles. Then, it 1 v1 ∼ exp exp(−iζx) as x → −∞. (11.10.7) v= v2 0 4ζ The form of the solution can be represented as a linear combination of two linearly independent solutions, that is, it 0 1 exp(iζx) + c12 exp(−iζx) as x → +∞. v ∼ exp c11 1 0 4ζ (11.10.8) Consequently, in the limit as x → +∞, t − ζx)], v1 ∼ c11 (ζ, t) exp[i( 4ζ
t v2 ∼ c12 (ζ, t) exp[i( 4ζ + ζx)],
(11.10.9ab)
where the time dependence of c11 and c12 is found from (11.10.2ab) in the limit as x → +∞. Since we consider the case where u → 2nπ as x → +∞, v1t ∼
i v1 , 4ζ
v2t ∼ −
i v2 4ζ
as x → +∞.
(11.10.10ab)
Following the method used for the modified KdV equation, it turns out that it , c12 (ζ, t) = c12 (ζ, 0). c11 (ζ, t) = c11 (ζ, 0) exp − 2ζ
610
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
Using the notation RR (ζ, t) = c11 (ζ, t)/c12 (ζ, t)
and mL (k , t) = −ic11 (k , t)/c12 (k , t),
we find the following result: t − ζz dζ RR (ζ, 0) exp −i 2ζ −∞ N t − mR (k , 0) exp −i − k z . 2k
1 ΩR (z, t) = 2π
∞
(11.10.11)
=1
Similarly, we can derive t − ζz dζ RL (ζ, 0) exp −i 2ζ −∞ N t − mL (k , 0) exp −i − k z . 2k
1 ΩL (z, t) = 2π
∞
(11.10.12)
=1
A procedure similar to that used for the modified KdV equation can be employed to determine the pure soliton solutions without any difficulty. Making reference to Lamb (1980), the multisoliton solutions are then given by −1 Im|I − iM | , (11.10.13) u(x, t) = −4 tan Re|I − iM | where I is a unit matrix of order N , N is the number of zeros of c12 (ζ, 0) in the upper half-plane, and M is an N × N matrix with time-dependent elements it . (11.10.14) MLj (kj , t) = mLj (kj , 0) exp 2kj Here, we consider the simplest case which deals with a single pole on the imaginary axis at k1 = 12 ia1 so that t mL1 (k1 , t) = mL1 (k1 , 0) exp (11.10.15a) a1 and ML1
1 t = mL1 (k1 , 0) exp a1 x + . a a1
Consequently, the single-soliton solution is derived from (11.10.13) as t +γ , u(x, t) = 4 tan−1 exp a1 x + a1
(11.10.15b)
(11.10.16)
where γ = log[ a1 · mL1 (k1 , 0)]. This solution is identical to (11.8.11) which was found earlier by a different, but simpler approach.
11.11 The Similarity Method for the Sine-Gordon Equation
611
Finally, the solution of the sine-Gordon equation representing the interaction of two solitons can also be obtained from two poles on the imaginary axis. This case is similar to that of the multisoliton solution of the KdV equation. Here, we set kr = 12 iar , r = 1, 2. Then the expression |I − iML | takes the form 2
mL1 mL2 a1 − a2 Re|I − iML | = 1 − exp (a1 + a2 )x , a1 a2 a1 + a 2 mL1 mL2 Im|I − iML | = − exp(a1 x) + exp(a2 x) . a1 a2 Consequently, the two-soliton solution takes the form a1 + a2 cosh 12 (u1 − u2 ) −1 u(x, t) = −4 tan , a1 − a2 sinh 12 (u1 + u2 ) where ur =
t ar x + + γr , ar
exp(γr ) =
a1 − a2 a1 + a2
a1 > a2 ,
(11.10.17) (11.10.18)
(11.10.19)
mLr (k, 0) . (11.10.20) ar
The above solutions are those special solutions which evolve into pure multisoliton solutions.
11.11 The Similarity Method for the Sine-Gordon Equation To obtain a similarity solution for the characteristic form of the sine-Gordon equation (11.9.1), we assume that this equation is invariant under a group of transformations x ˜ = aα x,
t˜ = aβ t,
and u ˜ = aγ u,
for suitable values of constants α, β, and γ. It follows that
u ˜x˜t˜ − sin u ˜ = aγ−α−β uxt − sin aγ u = uxt − sin u,
(11.11.1)
(11.11.2)
provided that γ = 0 and α = −β. This shows that equation (11.9.1) admits similarity transformations given by u(x, t) = v(η), ∂ d Using ∂x = t dη , equation
∂ ∂t
η = xt−α/β = xt.
(11.11.3)
d = x dη , equation (11.9.1) reduces to the ordinary differential
ηv (η) + v (η) = sin v,
(11.11.4)
where the prime denotes differentiation with respect to η. We next introduce a new dependent variable w = exp(iv) in (11.11.4) to derive an equation for w in the form
612
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
Fig. 11.5 Similarity solution of (11.11.4). From Lamb (1980).
w −
1 w2 + 2w − w 2 + 1 = 0. w 2η
(11.11.5)
Equation (11.11.5) is a special case of the equation that defines the Painlevé equation of the third kind (Davis 1962, p. 185). However, it is normally not easy to find a solution of (11.11.5); hence, it is necessary to use numerical integration to do so. An example of such a solution is given by Lamb (1980) with a phase plane analysis = utx by (11.11.3). and graphical representations of functions v and v = − 2q t The numerical solutions for v and v shown in Figure 11.5 satisfy the conditions v(0) = 0.1 and v (0) = sin v(0), giving a solution that is finite at η = 0.
11.12 Nonlinear Optics and the Sine-Gordon Equation McCall and Hahn (1967, 1969) discovered a remarkable new wave mode which propagates in a two-level atomic system without attenuation. When a sufficiently intense electromagnetic wave is incident on the system as a short pulse, then the number of atoms in the excited state in the leading part of the pulse exceeds that in the ground state due to the absorption of the electromagnetic wave. However, the electromagnetic wave in the rear part of the pulse is again emitted due to induced emission so that the system transmits a pulse without absorption. This is called self-induced transparency (SIT).
11.12 Nonlinear Optics and the Sine-Gordon Equation
613
In SIT, a medium transmits a light pulse at the resonant frequency without resonance absorption. When an electromagnetic wave is incident on a two-level atomic system, an ensemble of atoms has a ground state of energy E1 and an excited state of energy E2 . If the frequency of the wave is equal to the transition frequency ω0 = (E2 − E1 )/, the transition from the ground state to the excited state occurs due to the resonance absorption of the electromagnetic wave. If the present system is composed of a large number of atoms, the atoms excited to the higher level lose energy by collision with atoms and the slow irreversible decay of the number of excited atoms to lower levels. However, the remarkable fact is that the pulse propagates as a soliton without attenuation of energy. To describe the self-induced transparency phenomena, we may regard the electromagnetic wave as a modulated, circularly polarized plane wave traveling along the x-axis so that the electric field E(x, t) is represented by
ˆ sin θ , E(x, t) = E(x, t) ˆj cos θ + k (11.12.1) where θ = (kx − ωt) + ψ(x), k = nω c , is the refractive index determined by the linear dispersion relation of the medium, the amplitude E(x, t) and the phase ψ(x) vary slowly compared with that of the carrier wave, that is, ∂E ∂E |ωE|, |kE|, (11.12.2) ∂t ∂x and similar results hold for ψ(x). The electric field E(x, t) can be obtained from the Maxwell equations and satisfies the wave equation 2 2 2 ∂2E n ∂ E 4π ∂ P − = , (11.12.3) ∂x2 c ∂t2 c2 ∂t2 where P(x, t) is the polarization of the medium caused by the electromagnetic wave. If the equation for P(x, t) is given, this equation combined with (11.12.1) forms a closed system for two unknowns E and ψ. We assume that the medium is made up of N two-level atoms and the transition frequencies ω0 of individual atoms are not all the same but are distributed about the frequency ω0 of the incident wave field in accordance with the spectral density function f (Δω), which is normalized by the conditions ∞ Δω = (ω0 − ω) and f (Δω) d(Δω) = 1. (11.12.4) −∞
Consequently, the polarization of the medium reduces to ∞ f (Δω)p(Δω, x, t) d(Δω), P(x, t) = N
(11.12.5)
−∞
where p(Δω, x, t) is the polarization of a two-level atom induced by the electromagnetic wave and p(Δω, x, t) satisfies the slow-variation conditions of (11.12.2).
614
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
Thus, when the polarization p consists of two components, one is in phase with the carrier wave of the electric field E(x, t) and the other is out of phase, so that it can be expressed as
ˆ sin θ p(Δω, x, t) = u(Δω, x, t) ˆj cos θ + k
ˆ cos θ , + v(Δω, x, t) ˆj sin θ − k (11.12.6) where u and v correspond to the modulations of the electric field and become slowly varying functions of x and t. Therefore, the problem reduces to determining the equations for u and v, which can be obtained from the Schrödinger equation for a two-level atom in the incident electric field. Invoking the electric dipole approximation and neglecting relaxation effects of the excited states, McCall and Hahn (1969) derived the following equations for u and v: ∂u = vΔω, ∂t 2 ∂v κ = −uΔω − EW, ∂t ω ∂W = vEω, ∂t
(11.12.7) (11.12.8) (11.12.9)
where u and v are usually called the dispersive and absorptive components of the electric dipole, respectively, κ = (2p/), p is the magnitude of the electric dipole moment, and W is the energy of the two-level atom, that is, W = − 12 ω0 in the ground state and W = 12 ω in the excited state. Moreover, we assume that all twolevel atoms are in the ground state until the dipole transitions occur. Thus, equations (11.12.3) and (11.12.7)–(11.12.9) constitute a closed system for u, v, W , and E. Since u, v, and E are slowly varying functions corresponding to the modulation of p, we derive equations for the modulational quantities E and ψ from (11.12.3). Substituting (11.12.1), (11.12.5), and (11.12.6) in equation (11.12.3) and using the slow-variation conditions (11.12.2), we derive the following equations for E and ψ: ∂E n ∂E 2πωN ∞ vf (Δω) d(Δω), (11.12.10) + =− ∂x c ∂t nc −∞ ∂ψ 2πωN ∞ E uf (Δω) d(Δω). (11.12.11) = ∂x nc −∞ Thus, we conclude that five equations (11.12.7)–(11.12.11) for u, v, W , E, and ψ describe the self-induced transparency phenomena. In particular, if all the transition frequencies of N two-level atoms are identical, so that f (Δω) = δ(Δω) and the frequency ω of the electromagnetic wave is equal to that frequency (Δω = 0), equations (11.12.7)–(11.12.11) can be reduced to the sine-Gordon equation. To show this, we can set u ≡ 0 and ψ ≡ 0 because of (11.12.7) and (11.12.11). Consequently, the system of equations (11.12.8)–(11.12.10) assumes the form
11.13 Nonlinear Lattices and the Toda-Lattice Soliton
2 ∂v κ EW, =− ∂t ω0 ∂W = ω0 vE, ∂t n ∂E 2πω0 N ∂E + =− v. ∂x c ∂t nc Introducing a new function φ(x, t) by φ(x, t) = κ
615
(11.12.12) (11.12.13) (11.12.14)
t
−∞
E(x, τ ) dτ
(11.12.15)
from (11.12.12), (11.12.13), we can write v(x, t) and W (x, t) in the form
v(x, t) = p sin φ(x, t) , (11.12.16a)
(11.12.16b) W (x, t) = W0 cos φ(x, t) , where W0 = − ω2 0 , and we have used the boundary condition that W (x, t) = − 12 ω0 , as t → −∞, that is, all the two-level atoms are in the ground state. Using equations (11.12.15) and (11.12.16a), equation (11.12.14) reduces to the sineGordon equation 1 (11.12.17) φxx + φtt = −γ 2 sin φ, c where c = nc and γ 2 = (πN ω0 )/nc. Using a suitable transformation of variables, equation (11.12.17) can be reduced to the canonical form of the sine-Gordon equation, φxx − φtt = sin φ.
(11.12.18)
11.13 Nonlinear Lattices and the Toda-Lattice Soliton We follow Toda (1967) and write the Toda lattice equation (2.3.36) with unit mass (m = 1) as the first-order system dsn = ae−brn − a = f (r), dt drn = 2sn − sn+1 − sn−2 . dt We note that s¨n = −abe so that
−brn
(11.13.1) (11.13.2)
drn dsn drn = −b a + dt dt dt
s¨n = sn+1 − 2sn − 2sn−1 , b(a + s˙ n )
where the dot denotes the derivative with respect to time t.
(11.13.3)
616
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
We next seek a traveling wave solution θ = ωt − αn
sn (t) = f (θ),
(11.13.4)
so that f (θ) satisfies the equation ω 2 f = f (θ + α) − 2f (θ) + f (θ − α). b(a + ωf )
(11.13.5)
In order to find f (θ), Toda suggested the identity involving elliptic functions d sn θ cn θ dn θsn 2 α dn 2 (θ + α) − dn 2 (θ − α) = −2k 2 , (11.13.6) dα 1 − k2 sn 2 θ sn 2 α where k is the modulus of the Jacobi elliptic functions. Using the Jacobi Epsilon function defined by θ E(θ) = dn 2 x dx, (11.13.7) 0
equation (11.13.6) is integrated to obtain E(θ + α) − 2E(θ) + E(θ − α) = −2k2
sn θ cn θ dn θ sn 2 α . (1 − k2 sn 2 θ sn 2 α)
(11.13.8)
This is close to equation (11.13.5). We further note that E (θ) = dn 2 θ = 1 − k sn 2 θ and E (θ) = −2k2 snθ cn θ dn θ so that (11.13.8) can be written as 2
E (θ) = E(θ + α) − 2E(θ) + E(θ − α), β + E (θ)
(11.13.9)
where β = sn12 α −1. The function E(θ) is not a periodic function, but it is associated with the periodic Jacobi zeta function Z(θ) ≡ E(θ) −
E(k) θ, K(k)
(11.13.10)
where K(k) = K and E(k) = E are complete elliptic integrals of the first and second kind, respectively, and 2K is the period of Z(θ). We next normalize the phase so a period corresponds to a unit increase in θ and write s˙n (t) in the form
E (2Kω)2 dn 2 2K(ωt − αn); k − . (11.13.11) s˙ n (t) = b K Thus, it follows from (11.13.1)–(11.13.2) that s˙ n 1 rn (t) = − log 1 + . b a
(11.13.12)
11.13 Nonlinear Lattices and the Toda-Lattice Soliton
617
In the limit as k → 1, E(k) → 1 and K(k) → ∞ and dn(z, k) → sech(z) so that the solitary wave solution is √
1 (11.13.13) rn (t) = − log 1 + sinh2 κ sech2 κn ± t ab sinh κ , b where κ is defined in such a way that 2Kα → κ,
2Kω →
√ ab sinh κ.
(11.13.14)
It follows from (11.13.13) that the velocity of the localized lattice wave can be expressed in terms of the amplitude parameter κ as √ sinh κ . (11.13.15) v = ab κ The above analysis of the Toda lattice problem leads to the following conclusions: 1. The solution (11.13.13) represents the Toda lattice soliton. The minus sign in (11.13.13) indicates that the Toda lattice soliton (TLS) is a compression wave. 2. The amplitude of the TLS tends to zero as sinh κ → 0 so that it reduces to a √ solution of the linear equation (2.3.37) moving with the velocity v = ab. 3. The velocity of a TLS is greater than that of small amplitude waves. 4. All these results of this analysis hold for all forms of the Toda potential from the linear limit of (2.3.37) to the hard sphere limit as b → ∞, while ab remains finite. 5. It also follows from (2.3.33)–(2.3.34) that Toda’s infinite chain of unit masses linked by the potential (2.3.35) is governed by the differential equation (2.3.34). The transformation
1 rn (t) = − log 1 + Vn (t) b reveals than Vn satisfies the equation
(11.13.16)
d2 log(1 + Vn ) = ab(Vn+1 − 2Vn + Vn−1 ). dt2
(11.13.17)
This represents that propagation of voltage through a nonlinear lattice filter. Intro√ ducing the time scale τ = t ab, Hirota (1973a, 1973b) showed that Vn (t) = where fn (τ ) =
μ=0,1
exp
(N ) i<j
d2 fn (τ ), dτ 2
Bij μi μj +
N
(11.13.18) μi Xi ,
i=1
Xi = ωi τ − κi n + γi , κi , κi , γi are constants, ωi = ±2 sinh 2 (ωi − ωj )2 − 4 sinh2 [(κi − κj )/2] exp(Bij ) = − . (ωi + ωj )2 − 4 sinh2 [(κi + κj )/2]
(11.13.19)
618
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
The results show that collisions of Toda solitons can be described by the N soliton solutions given by (11.13.18) and (11.13.19). In 1970s, it has been realized that a mass–spring lattice is completely integrable for the special form of the interaction potential (2.3.35) discovered by Toda (1967a, 1967b). However, even though exact integrability is lost for any other potential, it is still possible to obtain solitary wave solution of equation (2.3.34) in the form rn (t) = R(n − vt) = R(z),
(11.13.20)
with R(z) → 0 as |z| → ∞ so that equation (2.3.34) reduces to the form (11.13.21) v 2 R (z) = V R(z + 1) − 2V R(z) + V R(z − 1) . This is a general equation for long lattice solitary waves, where R(z) = rn (t) is the separation of adjacent lattice sites and V [R(z)] is the associated spring potential. Several authors have suggested methods for finding numerical or analytical approximation for lattice solitary waves (LSW). Eilbeck and Flesch (1990) have shown how to obtain fairly accurate numerical solutions of such equations (11.13.21) based on pseudospectral methods. They found R(z) as a finite Fourier cosine series of n terms over a finite interval. On the other hand, Duncan et al. (1993) have developed a rather general approach based on Fourier transform analysis. They defined the Fourier transform of F (z) = V [R(z)] by ∞
F˜ (k) = F F (z) = e−ikz F (z) dz, (11.13.22) −∞
˜ and similarly, R(k) = F{R(z)}. Thus, the Fourier transform of (11.13.21) is −1 4 k2 2˜ 2 k ˜ F˜ (k). (11.13.23) v R(k) = F (k) ≈ 1 + sin k2 2 12 This leads to the differential equation 1 d2 R(z) = V R(z) . v2 1 − 2 12 dz
(11.13.24)
Or equivalently,
12 2 v R(z) − V R(z) . (11.13.25) 2 v This equation can be integrated in terms of elliptic or hyperbolic functions for different forms of the function V [R(z)]. For a potential of the form R (z) =
V (r) =
1 2 ar3 r + , 2 3
(11.13.26)
Collins (1981) has obtained an approximate inversion of the different operator on the right-hand side of equation (11.13.21). We follow Duncan et al. (1993) with potential (11.13.26) to find equation (11.13.25) in the form
11.14 Exercises
R (z) =
12 2 v − 1 R(z) − 2 v
12a R2 . v2
619
(11.13.27)
We use a method similar to that of the KdV equation in Section 9.4 to find the solitary wave solution of (11.13.27) in the form
3 2 v − 1 sech2 (n − vt) 3 v 2 − 1 /v 2 . (11.13.28) R(z) = rn (t) = 2a Thus, this LSW can be either compressive for a < 0 or expansive for a > 0. In both cases, the velocity of the LSW is greater than unity, the wave velocity in the lowamplitude limit. In the limit as v 2 → 1+ with a < 0, the solution (11.13.28) tends to the Toda lattice soliton (11.13.13). On the other hand, as v 2 → 1+, the amplitude of the LSW (11.13.28) tends to zero so that its width increases without bound. This means that (11.13.28) represents a large solitary waves that extend over many lattice spacings. For more information about the existence of solitary waves on mass–spring lattices for a wider classes of potential functions and about the two-dimensional Toda lattice, the reader is referred to Scott (2003), Biondini and Wang (2010).
11.14 Exercises 1. Use the Fourier transform method to show that the Fourier transform solution of the (n + 1)-dimensional, inhomogeneous, Klein–Gordon equation 2 ∂2 ∂2 ∂2u ∂ 2 − c + + · · · + u + d2 u = q(x, t), x ∈ Rn , t > 0, ∂t2 ∂x21 ∂x22 ∂x2n with the Cauchy data u(x, 0) = f (x) is
and ut (x, 0) = g(x)
for all x ∈ Rn
U (k, t) = F (k) cos t c2 |k|2 + d2
G(k) + sin t c2 |k|2 + d2 c2 |k|2 + d2 t sin(τ c2 |k|2 + d2 ) Q(k, t − τ ) dτ, + c2 |k|2 + d2 0
where U (k, t) = F{u(x, t)}, x = (x1 , x2 , . . . , xn ), and k = (k1 , k2 , . . . , kn ). 2. Show that the energy integral associated with the Klein–Gordon equation in Exercise 1 is given by
2 E(t) = |ut | + c2 |∇u|2 + d2 |u|2 dx. Show that E(t) is independent of t.
620
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
3. Seek a similarity solution of the Klein–Gordon equation utt − uxx = u3 in the form u(x, t) = tm f (xtn ) for suitable values of m and n. Show that f (z) satisfies the equation 2
z − 1 f + 4zf + 2f = f 3 , where z = xt−1 . 4. If V (u) involved in the Klein–Gordon equation (11.5.1) is given (Whitham 1974) by 1 V (u) = u2 + bu4 + · · · , 2 derive the result 1 u(θ) = a cos θ + a3 b cos 3θ + · · · 8
as a → 0,
where the nonlinear dispersion relation is given by ω 2 = 1 + k2 + 3a2 b + · · ·
as a → 0
and the nonlinear amplitude is A= 5. Verify that
1 2 9 4 a + a b + ··· 2 8
as a → 0.
x − Ut u(x, t) = 4 tan−1 α exp √ 1 − U2
is a solution of the sine-Gordon equation (11.8.2), where α and U are constants. 6. Show that √ U cosh{x/ 1 − U 2 } √ u(x, t) = 4 tan−1 sinh{U t/ 1 − U 2 } is an exact antisoliton solution of the sine-Gordon equation, where 0 < |U | < 1. Examine the asymptotic nature of u(x, t), as t → ±∞, with physical significance. 7. Show that the transformations 1 t 1 t ξ= x− , η= x+ 2 c 2 c transform the equation uxx − c2 utt = sin u into the form uξη = sin u.
11.14 Exercises
621
8. (a) Show that the sine-Gordon equation uxx − utt = sin u is invariant under the transformations x ˜ = m(x − U t) and t˜ = m(x − U t), 2 − 12 where m = (1 − U ) and −1 < U < 1. (b) Hence or otherwise, derive that 1 sin{αβ(t − U x − t0 )} (1 − β 2 ) 2 u(x, t) = 4 tan−1 1 β cosh{α(1 − β 2 ) 2 (x − U t − x0 )} is a solution of the sine-Gordon equation, where β = 0, x0 , t0 , and U are real constants, −1 < β < 1, and −1 < U < 1. 9. (a) Use the Bäcklund transformation (11.9.2a), (11.9.2b) to show that four solutions ur (r = 1, 2, 3, 4) satisfy the relation 1 1 a1 + a2 tan (u4 − u1 ) = tan (u2 − u3 ) , 4 a1 − a2 4 where a1 and a2 are nonzero constants. (b) Apply the relation in (a) to obtain a soliton solution of the sine-Gordon equation by setting u1 = 0, and u2 and u3 as different soliton solutions. 10. Examine the asymptotic nature of the two-soliton solution (11.10.19), as t → ±∞. 11. Derive the solitary wave solution of the sinh-Gordon equation uxt = sinh u. 12. (a) Use the similarity solution of the sine-Gordon equation (11.9.1) of the form u(x, t) = f (xtn ) for suitable n to show that w(z) = exp{if (z)} yields a Painlevé equation for w(z). (b) Use the same method as in (a) to discuss the sinh-Gordon equation in Exercise 11. 13. Verify the following conservation laws (Lamb 1971) for equation (11.9.1): 1 2 1 2 − (1 − cos u)t = 0, (1 − cos u)x − ut ux = 0, 2 2 x t 1 6 2 2 2 8 3 4 2 1 4 4 2 + ut − ut utt + ut uttt + uttt ut − utt cos u = 0, 6 3 9 3 9 3 x t
1 4 2 2 + ut cos u t = 0. u − utt 4 t x 14. Using the Lagrangian density L=
1 2 ut − u2x − V (u), 2
where V (u) = sin u, derive the sine-Gordon equation uxx − utt = sin u.
622
11 Nonlinear Klein–Gordon and Sine-Gordon Equations
15. Show that the bilinear form of the sine-Gordon equation in Exercise 14 is the pair of equations
2
Dx − Dt2 − 1 (φ · ψ) = 0 and Dx2 − Dt2 (φ · φ − ψ · ψ) = 0, where u = 4 tan−1 ( ψφ ). 16. Consider the dissipative wave equation utt − c2 ∇2 u + νut = 0, where ν(> 0) is the coefficient of dissipation. Show that the energy is decreasing: dE = −ν u2t dx ≤ 0. dt 17. Use u(x, t) = exp(−at)v(x, t) to transform the one-dimensional dissipative wave equation in Exercise 16. For ν = a2 , show that the transformed equation is the Klein–Gordon equation with imaginary mass (ia). 18. Solve the fractional Klein–Gordon equation (Debnath and Bhatta 2004) ∂αu − c2 uxx + d2 u = p(x, t), x ∈ R, t > 0 (1 < α ≤ 2), ∂tα u(x, 0) = 0 = ut (x, 0), x ∈ R, u(x, t) → 0
as |x| → ∞, t > 0.
19. Solve the initial-value problem for the KG equation utt − c2 uxx + d2 u = 0, u(x, 0) = 0,
x ∈ R, t > 0, ut (x, 0) = δ(x),
x ∈ R.
12 Asymptotic Methods and Nonlinear Evolution Equations
The organic unity of mathematics is inherent in the nature of this science, for mathematics is the foundation of all exact knowledge of natural phenomena. David Hilbert It seems to be one of the fundamental features of nature that fundamental physics laws are described in terms of great beauty and power. As time goes on, it becomes increasingly evident that the rules that the mathematician finds interesting are the same as those that nature has chosen. Paul Dirac
12.1 Introduction Many physical systems involving nonlinear wave propagation include the effects of dispersion, dissipation, and/or the inhomogeneous property of the medium. The governing equations are usually derived from conservation laws. In simple cases, these equations are hyperbolic. However, in general, the physical processes involved are so complex that the governing equations are very complicated, and hence, are not integrable by analytic methods. So, special attention is given to seeking mathematical methods which lead to a less complicated problem, yet retain all of the important physical features. In recent years, several asymptotic methods have been developed for the derivation of the evolution equations which describe how some dynamical variables evolve in time and space. So, we begin this chapter with one simple method of construction of the linear evolution equation from a given frequency–wavenumber dispersion relation of the form ω = f (k). (12.1.1) This relation is multiplied by −iU (k) exp[i(kx − ωt)] and integrated with respect to the wavenumber k from −∞ to ∞ to obtain the equation L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_12,
624
12 Asymptotic Methods and Nonlinear Evolution Equations
∂u = L(u) = −i ∂t
∞
−∞
f (k)U (k) exp i(kx − ωt) dk,
(12.1.2)
where L(u) represents an operator and U (k) is an arbitrary function which is related to the function u(x, t) by the inverse Fourier transform ∞ U (k) exp i(kx − ωt) dk. (12.1.3) u(x, t) = −∞
Equation (12.1.2) represents the fundamental result that determines the structure of the evolution equation. In particular, if f (k) is a polynomial in k, then L(u) is a differential operator. More explicitly, if f (k) = ck − iγk2 − α3 k 3 − · · · − αn k n ,
(12.1.4)
where c, γ, α3 , . . . , αn are arbitrary constants, then (12.1.2) gives the explicit evolution equation ∂u ∂3u ∂ nu ∂u ∂ 2u +c − γ 2 + α3 3 + · · · + αn n = 0. ∂t ∂x ∂x ∂x ∂x
(12.1.5)
If α3 = α4 = · · · = αn = 0, (12.1.5) reduces to the linear Burgers equation. When αn ≡ 0 for n ≤ 4, equation (12.1.5) represents the linear KdV–Burgers equation. On the other hand, if γ = −iν for real ν and αn = 0 for n ≤ 3, then (12.1.5) reduces to the linear Schrödinger equation. The above analysis reveals that the polynomial dispersion relation corresponds to systems governed by evolution equations of the differential type. On the other hand, if the dispersion relation has a transcendental form, the corresponding evolution is of integro-differential type. However, there is no such simple method of construction of evolution equations from a given nonlinear dispersion relation. For nonlinear problems, asymptotic methods are employed to construct evolution equations. The linear part of the associated evolution equation can be recovered from (12.1.2). So, the rest of this chapter is devoted to deriving evolution equations for various nonlinear systems by asymptotic methods. The idea of a far-field asymptotic behavior of a physical system originated from the properties inherent in a given evolution equation which do not depend, in any sensitive manner, on the details of the initial conditions. Thus, such properties are generally observed after a sufficiently long period of time for a large class of initial conditions. In this sense, it is not truly useful to study the initial-value problems in detail. Rather, it is more useful to investigate the asymptotic behavior of a physical system in the limit as time goes to infinity. In many situations, these far-field solutions merely represent the simple waves described in Chapter 6. Included here are reductive perturbation methods for describing weakly dispersive systems, strongly dispersive systems, and quasi-linear dissipative systems. The perturbation method due to Ostrovsky and Pelinovsky (1971) is developed in Section 12.6 for a general system of first-order nonlinear partial differential equations
12.2 The Reductive Perturbation Method and Quasi-linear Hyperbolic Systems
625
that can be applied to both conservative and nonconservative systems. The Gardener– Morikawa transformation is introduced in Section 12.4 to derive a fairly general nonlinear evolution equation. This is followed by several examples of applications, including nonlinear oscillations of an elastic string with dispersive effects, the Burgers equation in gas dynamics, the KdV equation for ion-acoustic waves, and nonlinear shallow water waves on an uneven bottom. In Section 12.5, the reductive perturbation method is also applied to describe strongly nonlinear dispersive systems and the generalized nonlinear Schrödinger equation. The method of multiple scales and asymptotic analysis are introduced in Sections 12.7 and 12.8 and used to derive both the nonlinear Schrödinger and the KdV equations from nonlinear dispersive systems. Special attention is given to the asymptotic method that has been utilized for the derivation of the nonlinear Schrödinger equation and the Davey–Stewartson equations with direct applications to the instability of Stokes waves in water. Several conservation laws for the Davey–Stewartson equations are also derived in Section 12.9.
12.2 The Reductive Perturbation Method and Quasi-linear Hyperbolic Systems Historically, this method was first formulated by Taniuti and his collaborators (Taniuti and Wei 1968; Taniuti and Washimi 1968; Kakutani et al. 1968; Taniuti and Yajima 1969; Asano and Ono 1971; Taniuti 1974) in a more general form applicable to both weakly dispersive and weakly dissipative systems. We first consider a first-order, quasi-linear totally hyperbolic system of equations Ut + A(U )Ux = 0,
(12.2.1)
where U is a column vector with n components u1 , u2 , . . . , un and A is an n × n matrix. Such a system of equations has already been investigated in connection with simple waves. We expand the vector U about a constant state solution U (0) of (12.2.1) in terms of a small parameter ε which represents the smallness of the amplitude of U in the form U = U (0) + εU (1) + ε2 U (2) + · · · . (12.2.2) Since the amplitude is small, the solution of (12.2.1) can be approximated by the linearized form of the system in which A(U ) is replaced by A(U (0) ). As the system is totally hyperbolic, there are n real distinct eigenvalues λ(r) of A(U ), and hence, (r) there are n constant eigenvalues λ0 of A(U (0) ), where r = 1, 2, . . . , n. In the linear case, the characteristics are straight lines. On the other hand, for a nonlinear system, as time evolves, characteristics will no longer be parallel lines, and hence, after a sufficiently long time, characteristics intersect and lead to the breakdown of differentiability. To investigate the process more explicitly, we expand the jth family of characteristics of (12.2.1) in powers of ε, dx = λ(j) , dt
(12.2.3)
626
12 Asymptotic Methods and Nonlinear Evolution Equations
and, then, retain only terms of order ε to obtain (j) n dx ∂λ (j) =ε u(1) = εU (1) ∇u λ0 , r (j) dt (0) ∂ur U =U r=1
(12.2.4)
where x = x − tλ0 , so that the coordinate system moves with speed λ0 , λ0 = (1) λ(j) (U (0) ), ur is the rth element of the vector U (1) , and ∇u is the gradient operator (1) (1) (1) with respect to u1 , u2 , . . . , un . It follows from result (12.2.4) that the characteristics suffer from deviation from parallel straight lines by an amount of the order ε. Using the time scale εt, the phase (j) dx velocity d(εt) , moving with the speed λ0 in the reference frame, reduces to the order of unity. Hence, for small t ε−1 , the phase velocity can be well approximated by the phase velocity for the linear case. However, for large time scales, t ∼ ε−1 , the characteristics deviate significantly from those found in the linear case, even if the amplitude is very small (ε 1). Consequently, the far-field solutions can be described by using the stretched time scale t ∼ ε−1 . (j) It also follows from (12.2.4) that ∇u λ0 is constant, and hence, U is a function of x and εt. Consequently, the time derivative of U (1) is of the order ε and becomes small. This means that x is constant and U (1) varies slowly with time along the linear parallel equiphase straight lines. All of these lead to the use of the following transformations for the far-field solutions of (12.2.1): (j)
(j)
x = x − λ0 t, (j)
t = εt.
(j)
(12.2.5ab)
(j) ∂ . Hence, we use (12.2.1), (12.2.2) together ∂x
∂ ∂ ∂ ∂ = ∂x Clearly, ∂x and ∂t = ε ∂t −λ0 with the expansion of the matrix A as
A = A0 + εU (0) ∇u A0 + · · · ,
(12.2.6)
where A0 = A(U (0) ). Thus, we obtain the following system of perturbation equations: (j) (1) (12.2.7) O(ε) : λ0 I − A0 Ux = 0, (2) (1) (j) (1) (1) 2 O(ε ) : −λ0 I + A0 Ux + Ut + U ∇u A0 Ux = 0. (12.2.8) (1)
Equation (12.2.7) shows that Ux must be proportional to a right-eigenvector (j) (j) R0 of A0 corresponding to the eigenvalue λ0 . Integrating (12.2.7) with respect to x gives the solution (1) (j) (1) U (1) = φj R0 + Vj (t ), (12.2.9) where φj = φj (x , t ) is a function of x and t and Vj (t ) are vector functions of t only that arise from integration and so can be determined by the boundary (1) conditions. However, functions φj (x , t ) cannot be determined from (12.2.9), and we have to consider (12.2.8) to find them. Equation (12.2.8) is an algebraic equation (1)
(1)
(1)
12.2 The Reductive Perturbation Method and Quasi-linear Hyperbolic Systems (2)
(j)
627
(j)
for Ux , with a multiplying factor (−λ0 I + A0 ). If λ0 is an eigenvalue of A0 , the eigenvalue equation is given by
A0 − λ(j) I = 0. (12.2.10) 0 (j)
(j)
If L0 is a left-eigenvector of A0 corresponding to the eigenvalue λ0 , then (j)
L0
(j) −λ0 I + A0 = 0.
(12.2.11)
Thus, a necessary and sufficient condition for the existence of the solution of (12.2.8) is (j) (1) (1) L0 Ut + (U (1) ∇u A0 )Ux = 0. (12.2.12) Without loss of generality, we set Vj (t ) ≡ 0 in (12.2.9) so that the resulting equation (12.2.9) can be substituted in (12.2.12) to obtain the nonlinear partial (1) differential equation for φj (x , t ) in the form (1)
∂ (1) ∂ (1) φj + αj φj = 0, ∂t ∂x
(12.2.13)
where αj is given by (j)
αj = L 0
(j) (j) (j) (j) R0 ∇u A0 R0 /L0 R0 ,
which, after some simplification, reduces to (j) (j) αj = ∇u λ0 R0 .
(12.2.14)
(12.2.15)
Thus, the present asymptotic analysis allows us to derive (12.2.13), which represents the basic scalar evolution equation describing the far-field behavior of the jth (column vector) mode of wave propagation of the hyperbolic system (12.2.1). Further, to the order of ε, the solution of (12.2.1) can be written as (1)
(j)
U = U (j) + εφj R0 ,
(12.2.16)
(1)
where φj is a solution of (12.2.13). Evidently, the characteristic equation of (12.2.13) is dx (1) = αj φj , dt
(12.2.17)
which ensures the existence of the far-field behavior, provided that αj = 0. The solution (12.2.16) represents simple waves already discussed in Chapter 6 that describe the far-field behavior of (12.2.1) until a discontinuity is developed in the form of shock waves. The present method of analysis is called the reductive perturbation method because it reduces the determination of the far-field behavior of a system to the solution of a scalar, nonlinear evolution equation, such as (12.2.13).
628
12 Asymptotic Methods and Nonlinear Evolution Equations
Finally, we close this section by adding a remark on the nonuniqueness of the far-field solution of hyperbolic equations. Equation (12.2.13) is invariant under the transformation (1) = εφ(1) . t˜ = ε−1 t , φ (12.2.18) x ˜ = x , However, this transformation is not unique because equation (12.2.13) is also invariant under the following transformation: x ˜ = ε−a x,
(1) = εφ(1) , φ
t˜ = ε−(a+1) t ,
(12.2.19)
where a is an arbitrary constant. This transformation can be found from (12.2.5ab), and the corresponding transformation ξ = εa (x − λ0 t),
τ = ε(a+1) t,
(12.2.20)
leads to equation (12.2.13). This transformation was first introduced by Gardner and Morikawa (1960) for a solitary wave in a plasma propagating normally across a magnetic field and is known as the Gardner–Morikawa transformation. Since a is arbitrary, solutions vary more slowly over space and time for large values of a than they do for smaller values. Thus, the far-field solutions are not determined uniquely for hyperbolic equations. The breaking time tB after which smooth solutions cease to exist is determined by the initial conditions, so that, in the case of hyperbolic systems, the time scale for the far-field solution must be less than tB . Thus, the concept of a far-field solution for the hyperbolic case is not meaningful unless tB is sufficiently large. It is possible that a discontinuity (or shock) is developed before an initial disturbance breaks up into simple waves. The above asymptotic method can be applied to physical systems with dissipative terms which prevent breaking. So, smooth solutions exist for all time for dissipating systems, so that tB → ∞, unlike the case of hyperbolic equations. This leads us to consider an arbitrarily long time evolution of a solution. For example, in the case of a perfect gas, the viscosity begins to act effectively as the sound wave steepens, and these two effects, the nonlinearity leading to breaking and the dissipation that causes smoothing, balance one another to produce smooth solutions. As a result, shock waves are developed and propagated as already described by the Burgers equation in Chapter 8.
12.3 Quasi-linear Dissipative Systems We now apply the reductive perturbation method to solve a general dissipative system which becomes totally hyperbolic when the dissipative terms are neglected. Using the same notation as in (12.2.1), we consider the system of dissipative equations described by Ut + A(U )Ux + K1 (K2 Ux )x = 0, (12.3.1) where K1 and K2 are n×n matrices which are functions of U and the eigenvalues of A(U ) are all real and distinct because of the first-order hyperbolic system embedded in the system of equations (12.3.1).
12.3 Quasi-linear Dissipative Systems
629
We now adopt a frame of reference that moves with a constant velocity λ0 which is equal to one of the eigenvalues of the matrix A0 = A(U (0) ) associated with the constant state U (0) . We assume that the dispersion relation for small wavenumbers k of the linearized system about U (0) is given by ω = λ0 k + iμk 2 + · · · ,
(12.3.2)
where μ < 0. It can be anticipated that equation (12.3.1) can be reduced to the Burgers equation by the Gardner–Morikawa transformation (12.2.20) with a = 1. In view of this transformation, equation (12.3.1) admits approximate solutions which vary slowly over space and more slowly with time due to weak nonlinearity and weak dissipating effects. With ξ = ε(x − λ0 t) and τ = ε2 t, we apply the same perturbation solution (12.2.2) and use the following results: 2 ∂2 2 ∂ = ε , ∂x2 ∂ξ 2
∂ ∂ =ε , ∂x ∂ξ
∂ ∂ ∂ = −ελ0 +ε , ∂t ∂ξ ∂τ
(12.3.3)
in the original equation (12.3.1). Then, equating the lowest-order terms O(ε2 ) gives (1)
(λ0 I − A0 )Uξ
= 0,
(12.3.4)
U (1) = φ(1) (ξ, τ )R0 ,
(12.3.5)
so that integration of this result gives
where the scalar function φ(1) is to be determined, R0 is a right-eigenvector of A0 corresponding to the eigenvalue λ0 , and the boundary condition U (1) → 0, as ξ → ∞, has been utilized. It is noted that (12.3.5) is not affected by the dissipative term to this order of approximation. Using the same argument as before, we obtain the following equation to the order O(ε3 ): (2)
(−λ0 I + A0 )Uξ
(1)
+ Uτ(1) + U (1) ∇u A0 Uξ
(1)
+ K0 Uξξ = 0,
(12.3.6)
where K0 = (K1 K2 )U =U (0) . Multiplying from the left by a left-eigenvector L0 corresponding to the eigenvalue λ0 and using the same argument as in Section 12.2 yields the necessary and (2) sufficient condition for the existence of solutions for Uξ as (1) (1) L0 Uτ(1) + U (1) ∇u A0 Uξ + K0 Uξξ = 0.
(12.3.7)
Substituting U (1) from (12.3.5) in (12.3.7) gives the following quasi-linear, second-order equation for φ(1) (ξ, τ ) ≡ φ(ξ, τ ): φτ + αφφξ + μφξξ = 0, where
(12.3.8)
630
12 Asymptotic Methods and Nonlinear Evolution Equations
α = (∇u λ0 )R0 L0 K0 R0 . μ= L0 R0
and
(12.3.9a) (12.3.9b)
Thus, the scalar function φ(ξ, τ ) satisfies the Burgers equation (12.3.8), provided that μ < 0. This equation governs the far-field behavior of the system (12.3.1), corresponding to the characteristic field λ0 of the hyperbolic part of the system. Finally, the solution of system (12.3.1) to the order O(ε) is given by U = U (0) + εφ(1) (ξ, τ )R0 .
(12.3.10)
Thus, φ(1) (ξ, τ ) ≡ φ(ξ, τ ) satisfies a nonlinear evolution equation (12.3.8) which determines the far-field behavior corresponding to the characteristic field λ0 .
12.4 Weakly Nonlinear Dispersive Systems and the Korteweg–de Vries Equation The dispersion relation for nonlinear waves represents an algebraic equation for the frequency ω, wavenumber k, and amplitude a2 : ω = ω(k, a2 ). If the amplitude is small, the dispersion relation is usually determined by the linearized equation, so that both the phase and group velocities are functions of the wavenumber k only. The dispersion is said to be weak if the group velocity of the wave is almost constant in ∂2ω a certain range of frequencies, and such that | ∂ω ∂k | | ∂k2 |. The Korteweg–de Vries (KdV) equation is a classic example of weak dispersion. However, if this inequality is not valid, the dispersion is said to be strong. The nonlinear Schrödinger (NLS) equation is a famous example of strong dispersion. We develop the reductive perturbation method to solve a general weakly nonlinear dispersive system in a homogeneous medium given by s p β ∂ β ∂ Hα U = 0, p ≥ 2, (12.4.1) + Kα Ut + AUx + ∂t ∂x α=1 β=1
where U is a column vector of n components u1 , u2 , . . . , un (n ≥ 2), and A, Hαβ , and Kαβ are n × n matrices, which are functions of U . For suitable Hαβ and Kαβ , the nonlinear system can be dispersive because the highest spatial derivative is of order three, and it is weakly dispersive because it does not contain an (n×1) column vector of the form B(U ) on the left-hand side of (12.4.1). Further, we can take into account the nonhomogeneity of the medium by adding an extra term BVx on the left-hand side of equation (12.4.1). We first consider a solution of (12.4.1) representing small derivations around a steady equilibrium state U0 in the form (12.4.2) U = U0 + U1 exp i(kx − ωt) .
12.4 Weakly Nonlinear Dispersive Systems and the Korteweg–de Vries Equation
631
Substituting (12.4.2) in (12.4.1) and retaining only the first power of U1 gives p s ω ω β β (12.4.3) Kα0 − Hα0 U1 = 0, − I + A0 + (ik)p−1 k k α=1 β=1
where the zero subscript denotes the value when U = U0 . Thus, equation (12.4.3) leads to the dispersion relation
p s
ω β
ω
β p−1 (12.4.4) Kα0 − Hα0 = 0. I + A0 + (ik)
−
k k α=1 β=1
For the case of long waves, that is, for small k, equation (12.4.3) can be solved by the method of successive approximations. The zeroth-order approximation is given by ω − I + A0 U10 = 0. (12.4.5) k Obviously, the zeroth-order dispersion relation is of order n in ( ωk ). We assume that λ0 is a nondegenerate eigenvalue of A0 with L0 and R0 as corresponding left- and right- eigenvectors, respectively. To this approximation, we find ω = λ0 , k
U10 = R0 or some scalar multiple of R0 .
(12.4.6)
Then we obtain the next approximation by substituting λ0 for ωk and R0 for U1 in the neglected terms. Consequently, p s β ω β p−1 I + A0 + (ik) Kα0 − λ0 Hα0 R0 = 0, (12.4.7) − k α=1 β=1
so that the multiplication of this equation by L0 yields s p (ik)p−1 L0 β ω β = λ0 + Kα0 − λ0 Hα0 R0 . k L0 R0 α=1
(12.4.8)
β=1
Proceeding in this manner gives a higher-order approximation to ( ωk ) as ω = λ0 + C1 k p−1 + C2 k 2(p−1) + · · · , k where the coefficient C1 is given by s p β β −1 p−1 Kα0 − λ0 Hα0 R0 , C1 = (L0 R0 ) i L0 β=1 α=1
and similarly for C2 . In this system, it is necessary to assume that C1 = 0.
(12.4.9)
(12.4.10)
632
12 Asymptotic Methods and Nonlinear Evolution Equations
Neglecting the third term in the original equation (12.4.1), the characteristic curves of the reduced equation can be expressed in the form dx = λ0 + ελ1 + O ε2 , dt
(12.4.11)
where ε is a small nonzero parameter which determines the effects of nonlinearity. A simple comparison of (12.4.9) with (12.4.11) leads to the fact that there can be a coupling between nonlinearity and dispersion (or dissipation) at the order of ε, provided that k ∼ εa , a = (p − 1)−1 . (12.4.12) This means that εa times the wavelength is of order unity, so that we can define the moving coordinate ξ by (12.4.13) ξ = εa x − λ 0 t , in which the wave can be described when nonlinearity is weak and the wavelength is long. Differentiating (12.4.13) with respect to x gives −1 dx dξ a = ε 1 − λ0 dx dt λ0 a a+1 λ1 = ε 1− =ε + O(ε) , λ0 + ελ1 + O(ε2 ) λ0
(12.4.14)
or equivalently, λ1 dξ = + O(ε), dη λ0 η = εa+1 x. This allows us to introduce a stretched variable η by (12.4.15b) so that Similarly, differentiating (12.4.13) with respect to t gives dξ = εa+1 λ1 + O(ε) . dt
(12.4.15a) (12.4.15b) dξ dη
= O(1).
(12.4.16)
This enables us to define another stretched variable τ by τ = εa+1 t
(12.4.17)
dξ such that dτ = O(1). We thus obtain two sets of stretched variables: (i) ξ, η and (ii) ξ, τ . Using the first set defined by (12.4.13) and (12.4.15b) gives the characteristics in terms of these stretched variables in the form
dξ dx = λ0 + ελ0 . dt dη
(12.4.18)
The first set of stretched variables can be used to deal with an initial-value problem.
12.4 Weakly Nonlinear Dispersive Systems and the Korteweg–de Vries Equation
633
Similarly, the characteristics associated with the second set ξ, τ are given by dx dξ = λ0 + . dt dτ
(12.4.19)
The second set of stretched variables introduces the Gardner–Morikawa transformation and can usually be applied to deal with boundary-value problems. It is clear from the above analysis that the scale of coordinate stretching is uniquely determined by the given governing equation (12.4.1). The reduction perturbation analysis has been developed by Taniuti and Wei (1968) for the case of a homogeneous medium and, subsequently, modified by Asano and Ono (1971) to account for moderate nonhomogeneity. When the medium is not homogeneous, as is the case with most physical systems, the interaction of the wave with the nonhomogeneity also becomes important. To state the stretched variable transformations associated with a nonhomogeneous medium, we briefly outline the basic ideas of Asano and Ono (1971) as follows: The steady state of the system of (12.4.1) with the term BVx on the left-hand side, denoted by the zero subscript, is described by A0 U0x +
p s β ∂ U0 + B0 Vx = 0. Kα0 ∂x α=1
(12.4.20)
β=1
This arises from a more general system of equations for weakly nonlinear waves in a homogeneous medium. With the assumption that U0 and V are slowly varying functions of x, the stretched variable η can be defined by (12.4.15b) with a = (p − 1)−1 . We can also determine the order of ε from the slow variations of U0 and V with x. To account for the nonhomogeneity of the medium, we introduce the stretched variable ξ and η redefined by the following relations: λ−1 ξ = εa dx − t and η = εa+1 x, (12.4.21ab) 0 where λ0 is a nondegenerate eigenvalue of the matrix A0 , equal to the velocity of the linear wave. In terms of stretched coordinates, equation (12.4.20) becomes A0 U0η + B0 Vη = 0,
(12.4.22)
where the terms of the order εp are ignored. The reader is referred to Asano and Ono (1971) or Bhatnagar (1979) for a complete analysis. However, we simply state the final evolution equation for φ(ξ, η) in the form φη + (α1 φ + α2 )φξ + β1 φξξ...ξ + γ1 φ + γ2 = 0, where the partial derivatives in the third term of (12.4.23) occur p times,
(12.4.23)
634
12 Asymptotic Methods and Nonlinear Evolution Equations
U1 = R0 φ(ξ, η) + W (η), α1 = (λ20 L0 R0 )−1 L0 {(∇U · A)0 · R0 }R0 , α2 = (λ20 L0 R0 )−1 L0 {(∇U · A)0 · W }R0 , s p 1 β β −1 β1 = (λ0 L0 R0 ) L0 −Hα0 + Kα0 R0 , λ0 β=1 α=1 γ1 = (λ0 L0 R0 )−1 λ0 L0 R0 + L0 (∇U A)0 R0 U0η + L0 (∇U B)0 R0 Vη , γ2 = (λ0 L0 R0 )−1 λ0 L0 Wη + L0 (∇U A)0 W U0η + L0 (∇U B)0 W Vη , and W is an arbitrary function of η alone. Following Taniuti and Wei (1968), we outline the fundamental ideas of the reductive perturbation method for a systematic reduction of a fairly general nonlinear system of equations (12.4.1) to a single, tractable, nonlinear evolution equation describing the far-field behavior. We introduce the Gardner–Morikawa transformation ξ = εa (x − λt),
τ = εa+1 t,
a = (p − 1)−1 ,
(12.4.24)
and assume expansions about a constant solution U (0) in terms of the small parameter ε of the form U = U (0) + εU (1) + ε2 U (2) + · · · , ∞ ∞ β εj Aj , Hαβ = εj Hαj , A= j=0
j=0
(12.4.25) Kαβ =
∞
β εj Kαj . (12.4.26)
j=0
Substituting (12.4.25), (12.4.26) in (12.4.1) and using (12.4.24) enables us to rewrite (12.4.1) in terms of derivatives with respect to ξ and τ . Equating the coefficients of like powers in ε to zero yields ∂U (1) O εa+1 : (−λI + A0 ) = 0, (12.4.27) ∂ξ ∂U (1) ∂U (1) (1) ∂U (2) + + U · (∇U A)0 O εa+2 : (−λI + A0 ) ∂ξ ∂τ ∂ξ p s p (1) β β ∂ U + −λHα0 + Kα0 = 0, (12.4.28) ∂ξ p α=1 β=1
denotes the gradient operator with respect to U , (U0 · ∇U ) reprewhere, again, ∇U ∂ sents the operator ni=1 ui ∂u , and A1 is written as i U (1) · (∇U A)U =U (0) ≡ U (1) · (∇U A)0 . Introducing a right-eigenvector R0 of A0 corresponding to the eigenvalue λ, so that (A0 − λI)R0 = 0 and integrating (12.4.27) with the boundary conditions U (1) → 0, as ξ → ∞, gives
12.4 Weakly Nonlinear Dispersive Systems and the Korteweg–de Vries Equation
U (1) = φ(1) (ξ, τ )R0 + V1 (τ ),
635
(12.4.29)
where φ(1) is one of the components of U (1) , and V1 (τ ) is an arbitrary vector-valued function of τ to be determined from the given boundary condition for U (1) . Since U (1) → 0 as ξ → ∞, V1 (τ ) = 0, and the solution (12.4.29) becomes U (1) = φ(1) (ξ, τ )R0 .
(12.4.30)
∂ To solve for ∂ξ U (2) , a compatibility condition is required. To find this condition, we multiply (12.4.28) on the left by a left eigenvector L0 and use the fact that L0 (−λI + A0 ) = 0 to obtain
L0
∂U (1) ∂U (1) + L0 · U (1) · (∇U A)0 ∂τ ∂ξ p s p (1) β β ∂ U −λHα0 + Kα0 = 0. + L0 · ∂ξ p α=1
(12.4.31)
β=1
Substituting (12.4.30) in this compatibility condition and dropping the superscript (1) leads to the nonlinear evolution equation for φ(1) (ξ, τ ) of the form ∂φ ∂φ ∂pφ + α1 φ + β1 p = 0, ∂τ ∂ξ ∂ξ where α1 and β1 are constants defined by α1 = L0 · R0 (∇U A)0 /(L0 R0 ), β1 = L0 ·
s
β β + Kα0 −λHα0 R0 /(L0 R0 ).
(12.4.32)
(12.4.33)
p
(12.4.34)
β=1 α=1
When p = 3, equation (12.4.32) reduces to the KdV equation, whereas for p = 2, (12.4.32) becomes the Burgers equation. Thus, the far field of the general nonlinear system (12.4.1) associated with the λ characteristics field satisfies the nonlinear equation (12.4.32), which admits the solution to the order O(ε) in the form U = U (0) + εφ(ξ, τ )R0 .
(12.4.35)
Taniuti and Wei (1968) have shown that the reduction perturbation method can be extended to the exceptional case in which the eigenspace of A0 comprises invariant subspaces. Su and Gardner (1969) derived both the KdV equation and the Burgers equation by using a similar perturbation technique. Finally, we describe far-field behavior by the modified KdV (mKdV) equation. The above analysis reveals that if the characteristic field associated with the hyperbolic part of system (12.4.1) is exceptional (linearly degenerate), then the nonlinear term in the KdV equation (12.4.32) vanishes (α1 = 0), and hence, (12.4.32) becomes linear. Such a linear equation cannot describe inherently nonlinear, far-field behavior, and hence, it is necessary to introduce a different scaling to retain the effects of weak nonlinearity on the far-field behavior.
636
12 Asymptotic Methods and Nonlinear Evolution Equations
We consider the situation in which the KdV equation can be replaced by the modified KdV equation in the form ut + us ux + μuxxx = 0
(12.4.36)
as the appropriate far-field equation. This equation is invariant under the transformations x ˜ = εα x, t˜ = εβ t, u ˜ = εu, (12.4.37) provided that α = − 2s and β = − 3s 2 . The solution of (12.4.36) is exceptional with respect to the characteristics field related to its hyperbolic part if the solution is expanded about u = 0. This follows from the fact that λ = us , so (∇u λ)u=0 = (sus−1 )u=0 = 0. Using the standard argument, it can be shown that the solution of (12.4.32) is also invariant under the transformation (12.4.37). So, if the characteristic field corresponding to the eigenvalue λ of A(U ) in (12.4.1) is exceptional, we can expand about a constant state U (0) after the change of variables ξ = εs/2 (x − λ0 t) and τ = ε3s/2 t and, then, write U = U (0) + εU (1) + ε2 U (2) + · · · .
(12.4.38)
Following the same argument used before, it can be shown that far-field behavior can be described by the modified KdV equation φτ + α1 φs φξ + μφξξξ = 0,
(12.4.39)
where α1 = 0. A wide variety of physical problems in plasma physics, nonlinear optics, and solid and fluid dynamics have been investigated by the reductive perturbation method or its extensions. The reader is referred to extensive references cited in several books and review articles by Jeffrey and Kakutani (1972), Taniuti (1974), Jeffrey and Kawahara (1982), Jeffrey and Engelbrecht (1994), Asano (1970), Asano et al. (1969, 1970), Asano and Ono (1971), and Kakutani (1971a, 1971b) who have investigated nonlinear wave propagation problems in inhomogeneous media. Asano (1974) examined the effects of weak dissipation or instability on nonlinear waves. The perturbation method has been generalized to allow for wave modulation in a system of nonlinear integro-partial differential equations by Taniuti and Yajima (1973). On the other hand, Nozaki and Taniuti (1973) and Oikawa and Yajima (1973, 1974a, 1974b) have considered the nonlinear interaction of three monochromatic waves and the interactions of solitary waves and nonlinear modulating waves. In addition, we may mention that a similar reduction perturbation theory has been developed in other contexts, for example, in studying the behavior of the partial differential equations in the neighborhood of a critical point and in describing the complete history of a pulse with a curved wave front. Relevant references include Kulikovskii and Slobodkina (1967), Bhatnagar and Prasad (1971), and Prasad (1973, 1975). We next illustrate the method by examples.
12.4 Weakly Nonlinear Dispersive Systems and the Korteweg–de Vries Equation
637
Example 12.4.1 (Oscillations of Nonlinear Elastic String with Dispersive Effects). This example is taken from Taniuti and Nishihara (1983). The transverse displacement y(x, t) of such a string of infinite length is governed by the nonlinear equation ytt − c2 yxx − σ 2 yxxxx = 0,
(12.4.40)
where c2 = (1 + yx2 )−3/2 and σ is a real constant. Equation (12.4.40) can be linearized when c ∼ 1, and hence, it admits a plane wave solution y(x, t) = A exp[i(kx − ωt)], provided that the dispersion relation ω 2 = k2 − σ2 k4
(12.4.41)
is satisfied. For appropriate long waves (large wavelength or small wavenumber), ω2 > 0, and so ω(k) is real, which confirms that the waves are dispersive. We now write yt = u and yx = υ, so that equation (12.4.40) can be replaced by a system of equations like (12.4.1) which has the form Ut + AUx + KUxxx = 0, where
u , U= v
A=
0 −1
−c2 , 0
(12.4.42)
0 −σ 2 and K = . 0 0
The eigenvalues of the matrix A are λ(±) = ±c, and hence, it follows that λ± 0 = ±1, for oscillations about the equilibrium position in which u = 0, whereas L± 0 = 1 (±) (±1, 1) and R0 = (1, ±1)T = ∓1 , so that ∇u λ(±) = (0, 0). Consequently, (±)
[∇u λ(±) ]0 R0 = 0, which confirms that each of the two characteristic fields associated with (12.4.42) is exceptional. The appropriate far-field equation in the present case satisfies the modified KdV equation 3 σ2 φτ − φ2 φξ + φξξξ = 0. 4 2
(12.4.43)
When both dissipation and weak dispersion are present in a physical system, the far-field behavior is governed by the KdV–Burgers equation. This equation cannot be derived by the standard reductive perturbation method. However, the necessary generalization of the method was given by Cramer and Sen (1992), including the case in which α in (12.3.9a) can change sign in a continuous manner, as the equilibrium solution U (0) is varied. This has led to the systematic study of nonlinear systems with positive or negative nonlinear terms depending on α > 0 or α < 0. Example 12.4.2 (Gas Dynamics and the Burgers Equation). For an ideal gas (p = RρT ), the conservation laws for mass, momentum, and energy are given by the following equations: ∂ ∂ρ + (ρu) = 0, ∂t ∂x
(12.4.44)
638
12 Asymptotic Methods and Nonlinear Evolution Equations
∂u ∂u 1 ∂p ν ∂ 2 u = 0, (12.4.45) +u + − ∂t ∂x ρ ∂x ρ ∂x2 ∂p p ∂ρ ∂u ∂p κ(γ − 1) ∂ 1 ∂p − 2 + γp +u − ∂t ∂x ∂x R ∂x ρ ∂x ρ ∂x 2 ∂ ∂ u (12.4.46) (uux ) = 0, + ν(γ − 1) u 2 − ∂x ∂x where ρ is the density, p is the pressure, u is the flow velocity, ν is the kinematic viscosity, κ is the thermal conductivity, and R and γ are gas constants. This system of equations can be expressed in the matrix form equation (12.4.1) with s = p = 2 if U , A, and K are given as follows: ⎛ ⎞ ⎛ ⎞ ρ u ρ 0 U = ⎝u⎠ , A = ⎝ 0 u ρ−1 ⎠ , p 0 γp u ⎛ ⎞ 0 0 0 ⎠, 0 0 0 K11 = ⎝ κ(γ − 1)R−1 −ν(γ − 1) −κ(γ − 1)R−1 ⎛ ⎛ ⎞ ⎞ 0 0 0 p/ρ2 0 0 − νρ 0⎠ , u 0 ⎠, K12 = ⎝0 K21 = ⎝ 0 −1 0 0 ρ 0 ν(γ − 1)u 0 ⎛ ⎞ 1 0 0 K22 = ⎝0 1 0⎠ . 0 0 1 1
2 The eigenvalues of matrix A are u, u ± ( γp ρ ) so that they reduce to 0 and ±1 for the (0) −1 = (1, 0, γ ). The eigenvectors L0 and R0 for a wave moving constant state U with velocity λ0 = 1 are given by L0 = (0, 1, 1) and R0 = (1, 1, 1). This leads to the Burgers equation
∂ρ(1) ∂ρ(1) ∂ρ(1) = 0, + α1 ρ(1) + β1 ∂τ ∂ξ ∂ξ 2
(12.4.47)
where ρ(1) is the first-order term of the expansion ρ = 1 + ερ(1) + · · · , the boundary condition U → U (0) as x → ∞ has been assumed, and α1 =
1 (γ + 1) and 2
β1 =
1 (γR)−1 κ(γ − 1)2 + ν . 2
Example 12.4.3 (Ion-Acoustic Waves and the KdV Equation). The motion of cold ions in a hot electron gas with the Boltzmann distribution law is governed by the equations ∂ ∂ni + (ni ui ) = 0, ∂t ∂x
(12.4.48)
12.4 Weakly Nonlinear Dispersive Systems and the Korteweg–de Vries Equation
Te ∂ne = eE, ne ∂x ∂ui 1 ∂ ∂ui e E− + ui = (ni Ti ), ∂t ∂x mi mi ni ∂x ∂E = 4πe(ni − ne ), ∂x
639
(12.4.49) (12.4.50) (12.4.51)
where the subscripts i and e refer to quantities related to ions and electrons, respectively, nj (j = i, e) is the density, uj (j = i, e) is the flow velocity, Tj is the product of the Boltzmann constant and the temperature, e is the charge of the ions, −e is the charge of the electrons, and E is the electric field. To apply the reductive perturbation method to this system of ion-acoustic waves, we eliminate ni and E from equations (12.4.48)–(12.4.51) to obtain ∂ui 1 ∂ne ∂ui + ui + = 0, ∂t ∂x ne ∂x ∂ ∂ ∂ ∂ue ∂ ∂ne n−1 = 0. + (ne ni ) − + ui e ∂t ∂x ∂x ∂x ∂x ∂x
(12.4.52) (12.4.53)
This system of equations corresponds to system (12.4.1) with p = 3 and s = 1. The associated matrices U , A, H, and K are given by ne ui ne U = , A= , n−1 ui ni e −1 0 1 0 K1 = , H2 = , H1 = 0, 0 0 0 0 −1 0 ui 0 ne , H3 = 0, and K3 = . K2 = 0 0 0 0 For U (0) = obtain
1 0
, the eigenvalues of A0 are λ0 = ±1. For the case of λ0 = 1, we 1 , and ∇u λ0 = (0, 1). R0 = L0 = (1, 1) 1
Thus, from (12.4.33) and (12.4.34), it turns out that α1 = 1, and β1 = 12 , and the (1) density of electrons ne to the first-order approximation satisfies the KdV equation (1)
∂ne ∂τ
(1)
+ n(1) e
(1)
∂ne 1 ∂ 3 ne + ∂ξ 2 ∂ξ 3
= 0.
(12.4.54)
In thermodynamic equilibrium at a constant temperature, electrons must satisfy the Boltzmann distribution law ne = n0 exp( eφ Te ), where n0 is the number density of electrons in the undisturbed state and φ is the electrostatic potential associated with (1) the wave motions. Expanding exp( Teφe ), we can obtain ne = φ(1) , and hence, the soliton solution of the KdV equation agrees with the solution for φ(1) .
640
12 Asymptotic Methods and Nonlinear Evolution Equations
Example 12.4.4 (Nonlinear Shallow Water Waves on an Uneven Bottom). For long waves on a beach, Peregrine (1967) derived the following system of coupled equations: ∂ 1 ∂u + (u · ∇)u + ∇h + H 2 ∇(∇ · u) ∂t 6 ∂t 1 ∂ − H ∇ ∇ · (Hu) = 0, 2 ∂t ∂h + ∇ · (H + h)u = 0, ∂t
(12.4.55) (12.4.56)
where u is the horizontal velocity averaged over the vertical direction, h is the wave amplitude, and H is the depth of still water, which depends only on the horizontal coordinates. For the motion of water in the x-direction only, Peregrine’s equations become ht + uhx + (H + h)ux + uHx = 0, 1 1 ut + uux + hx − H(Hut )xx + H 2 utxx = 0. 2 6
(12.4.57) (12.4.58)
It is noted here that only the first-order x- and t-derivatives are involved in equation (12.4.57), whereas equation (12.4.58) contains the third-order derivatives in which the x-derivative is repeated twice, but the t-derivative occurs only once. These comments are useful for choosing the matrices Hαβ and Kαβ in the original equation (12.4.1). In this problem, p = 3 and equations (12.4.57), (12.4.58) can be expressed in the form (12.4.1) with an extra term BVx on its left side by choosing s = 1 and u h u H +h , , A= , V = H, B= U = 1 u 0 u 0 H H11 = H21 = 0, H31 = , 0 1 0 0 K11 = , K21 = I, K31 = 0. − 12 H 16 H 2 For the equilibrium state h = u = 0, we have the matrix √ 0 H A0 = , and hence, λ0 = ± H. 1 0 A left-eigenvector L0 and a right-eigenvector R0 are given by H . L0 = (1, λ0 ) and R0 = λ0 To write the nonlinear √ evolution equation (12.4.23), we calculate its coefficients α1 , β1 , and γ1 for λ0 = H as
12.5 Strongly Nonlinear Dispersive Systems and the NLS Equation
3 − 12 1√ H , β1 = H, 2 6 5 1 dH , and γ2 = 0. γ1 = 4 H dη
641
α1 =
Consequently, the KdV equation for ψ = H 5/4 φ assumes the form 3 −7/4 1√ ψψξ + H ψη + H ψξξξ = 0, 2 6
(12.4.59)
(12.4.60)
where the coefficients of this equation are functions of η only.
12.5 Strongly Nonlinear Dispersive Systems and the NLS Equation We consider a general, strongly nonlinear, dispersive system that can be described by the equation s p β ∂ β ∂ Ut + AUx + B + U = 0, (12.5.1) Hα + Kα ∂t ∂x α=1 β=1
where U is a column vector with the n components u1 , u2 , . . . , un (n ≥ 2), A, Hαβ , and Kαβ are n × n matrices, and B is a column vector of n components, all of which are functions of U . We make an important assumption that the linear dispersion relation of (12.5.1) admits strong dispersion because of the term B. For simplicity, we consider the problem of nonlinear wave modulation based on a system of equations without the last term in (12.5.1), so that (12.5.1) gives the first-order equation Ut + AUx + B = 0.
(12.5.2)
Many physical problems exhibiting strong dispersion are described by this equation. A constant equilibrium solution U = U0 of this equation is required to satisfy the algebraic equation (12.5.3) B(U0 ) = 0. We assume here that a constant solution U0 exists and that a plane wave solution of infinitesimal amplitude is superimposed on the constant state, so that U = U0 + (δUk ) exp i(kx − ωt) + c.c., (12.5.4) where c.c. stands for the complex conjugate of the preceding expression. Substituting (12.5.4) in (12.5.2) and linearizing with respect to δUk gives the linear dispersion relation (12.5.5) | ∓ iωI ± ikA0 + ∇u B0 | = 0, where I is the unit matrix, A0 and B0 are matrices evaluated at U = U0 , and ∇U is the gradient in U space. Relation (12.5.5) determines ω as a function of k, and
642
12 Asymptotic Methods and Nonlinear Evolution Equations
hence, we assume that ω is given by a single real root of (12.5.5). We then introduce a matrix Wm defined by Wm = ∓imωI ± imkA0 + ∇u B0 .
(12.5.6)
To consider the modulation of a linear plane wave, we introduce the slow variables ξ and τ as ξ = ε(x − λt) and τ = ε2 t, (12.5.7ab) where λ = ( dω dk ) is the group velocity. The main objective here is to investigate how such a plane wave is modulated by nonlinear effects. We assume that the solution of (12.5.2) is of the form U = U0 +
∞
εn Un ,
∞
Un =
Un(m) (ξ, τ ) exp im(kx − ωt) , (12.5.8ab)
m=−∞
n=1
(m)
where Un is assumed to be real, so that Un is equal to its complex conju(m)∗ gates Un . We consider modulation of the plane carrier wave with wavenumber k and fre(m) quency ω. We may set U1 = 0 except for m = ±1. Substituting (12.5.8ab) into (12.5.2) and using (12.5.7ab) leads to the following perturbation equations: (m)
O(ε) : Wm U1
= 0,
(m) O ε2 : Wm U2 + (−λI + A0 ) +
U 1 · ∇U A 0
∞
(12.5.9)
(m) ∂U1
∂ξ (r) irkU1
exp ir(kx − ωt)
(m)
r=−∞
1 + ∇U ∇U : U1 U1 (m) = 0, (12.5.10) 2 (m) (m) ∂U1 ∂U (m) O ε3 : Wm U3 + (−λI + A0 ) 2 + ∂ξ ∂τ (m) (m) ∂U1 (r) + U 1 · ∇U A 0 + U1 · ∇U A0 irkU2 Z (r) ∂ξ r (m) 1 (r) U1 U1 : ∇U ∇U A0 + irkU1 Z (r) 2 r (m) (r) + U2 · ∇U A0 irkU1 Z (r) + ∇U ∇U B0 : U1 U2 (m) r
1 + ∇U ∇U ∇U B0 : U1 U1 U1 (m) = 0, 6
(12.5.11)
12.5 Strongly Nonlinear Dispersive Systems and the NLS Equation
643
where ·(m) represents the coefficient of the mth harmonic and Z (r) ≡ exp ir(kx − ωt) . Thus, from (12.5.9), we obtain (m)
U1
= φ1 (ξ, τ )R
for |m| = 1,
(12.5.12)
where φ1 is a complex function of slow variables ξ and τ to be determined and R is a right-eigenvector of W1 corresponding to the eigenvalue λ so that (12.5.13)
W1 R = 0, whereas, for |m| = 1, (m)
U1
(12.5.14)
= 0.
The third and fourth terms in equation (12.5.10) result from self-interaction of the (1) fundamental harmonic U1 exp[i(kx−ωt)], and they do not vanish only for |m| = 2 and for the slow mode m = 0. This implies that the second harmonic and the slow mode exist for the second order of ε. Thus, for m = 1, we simplify (12.5.10) by using (12.5.12) and (12.5.14) to obtain the equation ∂φ1 (1) W1 U2 + (−λI + A0 )R = 0. (12.5.15) ∂ξ (1)
Since det W1 = 0, equation (12.5.15) admits a solution for U2 compatibility condition L(−λI + A0 )R = 0
provided that the (12.5.16)
is satisfied, where L is a left-eigenvector corresponding to λ, so that LW1 = 0. Differentiating (12.5.13) with respect to k and using (12.5.6) gives ∂R = 0. i(−λI + A0 )R + W1 ∂k
(12.5.17)
(12.5.18)
Multiplying this equation by L from the left and using (12.5.17) yields the compatibility condition (12.5.16). Then, it follows from (12.5.15) combined with (12.5.18) that (1) ∂R ∂φ (2) (2) , (12.5.19) U1 = φ (ξ, τ )R − i ∂ξ ∂k where φ(2) is a complex function of ξ and τ . (0) (±2) Solving equation (12.5.10) for U2 and U2 , when m = 0 and m = ±2, we obtain solutions in the forms
644
12 Asymptotic Methods and Nonlinear Evolution Equations (0)
U2
(2)
U2
= −W0−1 ik(∇U A0 · R∗ − c.c.)
2 1 + (∇U ∇U B0 : R∗ R + c.c.) φ(1) , 2 2 1 = −W2−1 ik(∇U A0 · R)R + ∇U ∇U B0 : RR φ(1) . 2
(12.5.20) (12.5.21)
The nonzero terms in the summation over r in the third-order equation (12.5.11) (±1) (±1) (0) (±2) are only U1 , U2 , U2 , and U2 . However, for m = 1, there is no contribu(±1) due to the fact that tion from the term U2 (1) (r) (1) (1) Z Z = δ(r) and Z (−1) Z (r) = δ(r − 2), and hence, we note that, for |m| = 1, the nonlinear terms in (12.5.11) do not contain φ(2) . If we multiply equation (12.5.11) for |m| = 1 by L from the left and use equations (12.5.12)–(12.5.14) and (12.5.19)–(12.5.21), we can eliminate all terms containing φ(2) because the first and the second terms vanish by equations (12.5.17) and (12.5.16). Consequently, all terms involving φ(2) vanish, and hence, we obtain the equation for φ(1) in the form i
2
∂φ(1) ∂ 2 φ(1) +p + q φ(1) φ(1) = 0, 2 ∂τ ∂ξ
(12.5.22)
where the coefficients p and q are given by ∂R 1 · L(−λI + A0 ) · , (12.5.23) (L · R) ∂k iL (2) (2) ik 2(∇u A0 · R∗ )R2 − ∇u A0 · R2 R∗ q=− (L · R) ! 1 (2) + ∇u A0 · R0 R + (∇u ∇u A0 : RR∗ )R − (∇u ∇u A0 : RR)R∗ 2 1 (2) (2) + ∇u ∇u B0 : RR0 + R∗ R2 + ∇u ∇u ∇u B0 : RR∗ R , (12.5.24) 2
p=
(2)
(2)
where R2 and R0 are constant column vectors defined by
2 2 (2) (2) (2) (2)
and U0 = R2 φ(1)
U2 = R2 φ(1) and are given by
1 = ik(∇u A0 · R)R + ∇u ∇u B0 : RR , 2 (2) R0 = − W0−1 ik (∇u A0 · R∗ )R − c.c. 1 ∗ + (∇u ∇u B0 : R R) + c.c. . 2
(2) R2
−W2−1
(12.5.25ab)
(12.5.26)
(12.5.27)
12.5 Strongly Nonlinear Dispersive Systems and the NLS Equation
645
We next show that p = 12 ω as follows. Differentiating (12.5.18) with respect to k gives ∂R ∂ 2R (12.5.28) + W1 2 = 0. −iω R + 2i(−λI + A0 ) ∂k ∂k This is multiplied by L from the left to obtain, by (12.5.17), ∂R = 0, (12.5.29) −iω (L · R) + 2iL(−λI + A0 ) · ∂k which gives p = 12 ω from (12.5.23). We then derive the generalized nonlinear Schrödinger equation 1 −iφτ + ω φξξ + q|φ|2 φ = 0, 2
(12.5.30)
where φ is a function of ξ = ε(x − ω t) and τ = ε2 t, and q is not necessarily real. In particular, when q is real, (12.5.30) is called the nonlinear Schrödinger equation. Asano (1974) obtained the most general form of this equation for the case of threedimensional modulation of nonlinear waves as 2 2 ∂ ω ∂ φ 1 + q|φ|2 φ + iγφ = 0, (12.5.31) iφt + 2 i,j ∂ki ∂kj ∂ξi ∂ξj where ki (i = 1, 2, 3) and ω denote the components of the wavevector and the fre∂ω quency of the plane carrier wave, respectively, and ξi = ε{xi − ( ∂k )t}. The last i term iγφ is involved to incorporate weak dissipation and/or unstable effects. Equation (12.5.30) can be solved exactly. The transformations 1 ω τ = T, 2
ξ = X,
12
q
and ψ =
φ ω
are used to put it into the canonical form i
∂ψ + ψXX + 2σ|ψ|2 ψ = 0, ∂T
(12.5.32)
where σ = sgn( ωq ). This canonical equation has been solved in Chapter 10 by the inverse scattering method. For σ = +1, the asymptotic solution of the initial-value problem for (12.5.32) represents a sequence of envelope solitons ψ(X, T ) = 2η sech 2η(X + 4νT − X0 ) × exp −2iνX − 4i ν 2 − η 2 − iφ0 (12.5.33) and radiation modes, where η and ν are parameters to be determined from the initial conditions.
646
12 Asymptotic Methods and Nonlinear Evolution Equations
12.6 The Perturbation Method of Ostrovsky and Pelinovsky Whitham’s theory of averaged Lagrangian principles has been a major contribution to the field of nonlinear dispersive waves. This theory is of great interest from the viewpoint of its mathematical generality and physical clarity. Based on his averaged Lagrangian principles, Whitham derived his fundamental equations for describing slowly varying nonlinear dispersive waves. This method is directly applicable to a conservative system for which the Lagrangian is known. However, the determination of the Lagrangian is not always an easy problem. Furthermore, Whitham’s theory cannot be applied readily to nonconservative dynamical systems. To extend Whitham’s theory, Ostrovsky and Pelinovsky (1971, 1972) have developed an asymptotic perturbation method for a general system of first-order nonlinear partial differential equations that can be applied to both conservative and nonconservative systems. They also considered a dynamical system which can be described by nonlinear partial differential equations of the Lagrangian type that include dissipation. Ostrovsky and Pelinovsky (1972) also showed that the equations of the first approximation can be deduced from the generalized Hamilton variational principle. This section deals with the perturbation method due to Ostrovsky and Pelinovsky. We consider a system of first-order nonlinear partial differential equations M (ξ, τ, u, ux , ut ) = εN (ξ, τ, u, ux , ut ),
(12.6.1)
where u(x, t) is an n-dimensional column vector function, operators M and N denote sets of nonlinear functions ξ = εx and τ = εt, ξ and x are three-dimensional vectors, and the parameter ε characterizes the deviation from the stationary and uniform state of the medium. We also assume that operators M and N are sufficiently smooth in terms of their arguments. For ε = 0 and for constants ξ and τ , the system of equations (12.6.1) is assumed to have stationary plane wave solutions in the form u = U (θ)
and θ = κ · x − ωt.
(12.6.2ab)
These solutions can be determined from the ordinary differential equations (12.6.3) M ξ, τ, U (θ), −ωUθ , κUθ = 0, where ξ and τ are assumed to be constants. It is noted that solutions of (12.6.3) depend on m(≤ n) arbitrary constants of integration θ0 , A = (A2 , A3 , . . . , Am ) and on the parameters ξ, τ , κ, and ω. We also assume that the frequency ω, the wavenumber vector κ, and constant A satisfy the dispersion relation ω = ω(κ, A),
(12.6.4)
and that they will be chosen so that u = U (θ) is a periodic function of θ with period 2π. The nonlinear wave profile is then determined from the basic equation (12.6.3) and may, in general, be different from a sinusoidal shape.
12.6 The Perturbation Method of Ostrovsky and Pelinovsky
647
We now look for a solution in the form of the asymptotic series u = U (θ, A, ξ, τ ) +
∞
εn u(n) (x, t),
n=1 ∞
θ = θ (0) (x, t, ξ, τ ) +
εn θ (n) (ξ, τ ),
(12.6.5) (12.6.6)
n=1
A(ξ, τ ) =
∞
εn A(n) (ξ, τ ),
(12.6.7)
n=0
κ(ξ, τ ) = θx ,
and
ω(ξ, τ ) = −θt
(12.6.8ab)
so that
∂κ ∂ω + = 0. ∂t ∂x We also expand operators M and N in an asymptotic series to obtain
(12.6.9)
M = M (0) (U, −ωUθ , κUθ ) ∂M (0) ∂M (0) (1) ∂M (0) (1) u + Uξ + u(1) + U +ε + u τ t x ∂U ∂Ux ∂Ut 2 +O ε , (12.6.10) 2 (0) (12.6.11) εN = εN (U, −ωUθ , κUθ ) + O ε , (0)
where each term of the form ( ∂M ∂U )U represents a product of a square matrix and a column vector. We substitute (12.6.5)–(12.6.11) in (12.6.1) and equate coefficients of similar powers of ε to obtain ∂ ∂ u(r) = Q(n) , (12.6.12) P θ, ξ, τ, , ∂t ∂x where n = 1, 2, 3, . . . , ∂M (0) ∂M (0) ∂ ∂M(0) · + + ∂U ∂Ut ∂t ∂Ux (0) ∂M ∂U ∂M (0) ≡ N (0) − · · − ∂Ut ∂τ ∂Ux
P ≡ Q(1)
∂ , ∂x ∂U , ∂ξ
·
(12.6.13) (12.6.14)
and similar expressions for Q(2) , Q(3) , . . . . To determine functions θ and A to any desired approximation, we have to solve a system of nonhomogeneous, linear, partial differential equations with periodic coefficients and nonhomogeneous terms, which are also periodic in θ. Using the following results: ∂ ∂ ∂ ∂ ≡κ and ≡ −ω (12.6.15ab) ∂x ∂θ ∂t ∂θ
648
12 Asymptotic Methods and Nonlinear Evolution Equations
in equation (12.6.12), the solution u(n) can be represented in the form θ (n) R−1 Q(n) dθ , u =R
(12.6.16)
0
where R is a matrix consisting of vectors of the system of solutions of the variational equations ∂ P θ, ξ, τ, ω, κ, φ = 0, (12.6.17) ∂θ and R−1 is the inverse of the matrix R. It should be pointed out that the set of m particular vectors is found in terms of U as follows: R1 =
∂U , ∂θ
and
Rs =
∂U , ∂As
s = 2, 3, . . . , m.
(12.6.18)
Clearly, R1 is a periodic function of θ, and Rs can be expressed in the form Rs =
∂U ∂U , + Bs θ ∂As ∂θ
(12.6.19)
where Bs is an arbitrary constant to be determined from the dependence of κ and ω on the amplitude A. The remaining (n − m) vectors can be expressed as R = exp(λ θ)f (θ),
(12.6.20)
where f (θ) is a periodic function with period 2π. We assume that the characteristic exponent λ has no positive real part and, if Reλ = 0, then, Imλ = 0, ±1, . . . . Substituting R and R−1 in (12.6.16) yields θ m θ (n) (n) (n) UAr Q dθ dθ Uθ Q + u = Uθ 0
+ +
m
θ
UA r
UAr Q(n) dθ
0
r=2 n
0
r=2
fs exp(λs θ)
θ
exp(−λs θ ) fs∗ Q(n) dθ . (12.6.21)
0
s=m+1
The last term of (12.6.21) is bounded for all θ due to the nature of λ. In order for u(n) to remain bounded as a function of θ, it must satisfy the orthogonality conditions 2π UAr Q(n) dθ = 0 and (12.6.22a)
0 2π
Uθ Q(n) dθ = 0, 0
for n = 1, 2, . . . and r = 2, 3, . . . , m.
(12.6.22b)
12.7 The Method of Multiple Scales
649
The equations of the first-order approximation can be obtained in a different form. We multiply equation (12.6.1) by UA , integrate over θ, and then, subtract equation (12.6.22a) from the resulting integral expression to obtain
2π
2π
N (ξ, τ, U, −ωUθ , κUθ )UAr dθ.
M (ξ, τ, U, Ux , Ut )UAr dθ = ε 0
0
(12.6.23) Similarly, multiplying (12.6.1) by Uθ , integrating over θ, and subtracting equation (12.6.22b) from the integrated result gives
2π
2π
N (ξ, τ, U, −ωUθ , κUθ )Uθ dθ. (12.6.24)
M (ξ, τ, U, Ux , Ut )Uθ dθ = ε 0
0
The three equations (12.6.9), (12.6.23), and (12.6.24) form a closed set for the unknown functions κ, ω, and A. It is necessary to solve the second-order approximation equations to determine the phase function θ of order O(ε). In their paper, Ostrovsky and Pelinovsky (1972) have shown that equations of the first-order approximation can be derived from the generalized Hamilton variational principle in the averaged form. They examined the system of equations ∂ ∂L ∂ ∂L ∂L + − = φ, (12.6.25) ∂t ∂ut ∂x ∂ux ∂u φ = φ(0) + εφ(1) + O ε2 , (12.6.26) where L is the Lagrangian density, φ is the density of the nonconservative term, u(x, t) is an n-dimensional vector function, and ε is a small parameter. We assume that L and φ are sufficiently smooth functions of u, ut , ux and also of slow variables ξ ≡ εx and τ = εt. Equation (12.6.25) can be derived from the generalized Hamilton variational principle (δL + δW ) dx dt = 0,
(12.6.27)
where δW = φδu. We conclude the perturbation analysis of Ostrovsky and Pelinovsky by adding a couple of comments. First, this analysis is valid for nonlinear systems with arbitrary, but not necessarily, weak nonlinearity. Second, the reader is referred to the work of Ostrovsky and Pelinovsky (1971, 1972) and Gorschkov et al. (1974) for further information.
12.7 The Method of Multiple Scales Multiple scales or, more precisely, scales of different orders arise in many physical problems because different physical effects usually manifest themselves over different length and time scales. Naturally, the concept of multiple scales is involved, ex-
650
12 Asymptotic Methods and Nonlinear Evolution Equations
plicitly or implicitly, in almost all asymptotic perturbation methods. Hence, it seems to have wide applicability to many problems that involve physical phenomena which occur in relation to various scales. Although the method of multiple scales covers a wide variety of perturbation procedures, the method will be formulated here in the context of nonlinear wave propagation. Historically, Sturrock (1957) first introduced the method of multiple scales or, more precisely, the derivative expansion method for an investigation of nonlinear effects in electron plasmas. Nonlinear effects are analyzed by the above perturbation procedure to study the incoherent interaction which is responsible for spectral decay, that is, for the redistribution of energy in wavenumber space—the breakdown of organized large-scale motion into disorganized small-scale motion. Subsequently, Sandri (1963, 1965, 1967) used the method of multiple scales that completely separates the different time scales exhibited by the evolution of a gas to provide the foundation of nonequilibrium statistical mechanics. Several other authors including Stuart (1960), Watson (1960), Hasimoto and Ono (1972), Frieman (1963), Nayfeh (1965a, 1965b, 1971, 1973), Nayfeh and Hassan (1971), and Kawahara (1973, 1975a, 1975b) successfully applied the derivative expansion method to solve different problems in fluid mechanics and plasma physics. These authors showed that a systematic application of the derivative expansion method can effectively take into account the full effect of amplitude modulation of a quasimonochromatic wave, and thereby can lead to the nonlinear Schrödinger equation. It has also been shown that the derivative expansion can be applied successfully to problems with long-wave approximation, which leads to the Korteweg–de Vries equation. We consider the nonlinear partial differential equation ∂ ∂ ∂ ∂ (12.7.1) , , λ u(x, t) = N , , μ u2 (x, t), x ∈ R, t > 0, L ∂t ∂x ∂t ∂x where L and N are differential operators and λ, μ are fixed parameters. We introduce the sets of independent variables x0 , x1 , x2 , . . . , xm and t0 , t1 , t2 , . . . , tm defined by (12.7.2ab) xn = εn x and tn = εn t, where ε is a small parameter characterizing the smallness of the associated terms. Consequently, the dependent variable u(x, t) can be regarded as a function of these new variables so that u(x, t) = u(x0 , x1 , . . . , xm , t0 , t1 , . . . , tm ). Since the method is called the derivative expansion method, it is appropriate to introduce expansions of the derivative operators: m ∂ ∂ = εn ∂x n=0 ∂xn
and
m ∂ ∂ = εn . ∂t n=0 ∂tn
Using these expansions in (12.7.1) leads to the following results:
(12.7.3ab)
12.7 The Method of Multiple Scales
651
∂ ∂ ∂ ∂ ∂ ∂ ε Ln , ,..., , , ,..., ,λ = ∂t0 ∂t1 ∂tm ∂x0 ∂x1 ∂xm n=0 + O εm+1 , (12.7.4ab) m ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ n , ,μ = N ε Nn , ,..., , , ,..., ,μ ∂t ∂x ∂t0 ∂t1 ∂tm ∂x0 ∂x1 ∂xm n=0 + O εm+1 . ∂ ∂ L , ,λ ∂t ∂x
m
n
We also assume that u(x, t) has the asymptotic representation u(x0 , x1 , . . . , xm , t0 , t1 , . . . , tm ) m εn un (x0 , x1 , . . . , xm , t0 , t1 , . . . , tm ) + O εm+1 . =
(12.7.5)
n=1
In general, u(x, t) can be expanded in terms of another small parameter δ which measures the degree of nonlinearity of the wave field, and it is assumed that the new parameter is related to ε. For simplicity, however, u(x, t) has been expanded in powers of ε. We next substitute (12.7.4ab) and (12.7.5) in equation (12.7.1) and then equate coefficients of like powers of ε to obtain a system of perturbation equations from which it is possible to determine the functions un successively. The underlying assumption is that each perturbed quantity un must be nonsecular (bounded) at each stage of the perturbation process. Thus, the method of multiple scales can be applied effectively to a general dispersive wave system with or without small dissipation. In the rest of this section, we follow Jeffrey and Kawahara (1982) to illustrate the general method of multiple scales by applying it to the nonlinear Boussinesq equation utt − c2 uxx − λuxxtt =
1 2 u xx , 2
(12.7.6)
where c and λ are fixed constants. The dispersion relation for the linearized Boussinesq equation is given by D(ω, k) ≡ c2 k 2 − ω 2 − λk 2 ω 2 = 0.
(12.7.7)
Comparing the Boussinesq equation (12.7.6) with (12.7.1) yields L≡
2 ∂4 ∂2 2 ∂ − c − λ , ∂t2 ∂x2 ∂x2 ∂t2
N≡
1 ∂2 , 2 ∂x2
so that equation (12.7.6) can be written symbolically as L[u] = N u2 .
(12.7.8ab)
(12.7.9)
In view of the power series expansions (12.7.4ab), the first few operators are given by
652
12 Asymptotic Methods and Nonlinear Evolution Equations
2 ∂2 ∂4 2 ∂ , 2 −c 2 −λ ∂t0 ∂x0 ∂x20 ∂t20 ∂2 ∂2 ∂4 ∂4 L1 ≡ 2 , − 2c2 − 2λ + 2 ∂t0 ∂t1 ∂x0 ∂x1 ∂t0 ∂t1 ∂x20 ∂t20 ∂x0 ∂x1 2 ∂2 ∂2 ∂2 ∂4 ∂ ∂4 L2 ≡ 2 + 2 − c2 + 2 +2 2 − λ 2 2 2 ∂t1 ∂t0 ∂t2 ∂x1 ∂x0 ∂x2 ∂t ∂x ∂t0 ∂x0 ∂x2 0 1 4 4 4 ∂ ∂ ∂ +4 + 2 2 +2 , ∂t0 ∂t1 ∂x0 ∂x1 ∂t1 ∂x0 ∂t0 ∂t2 ∂x20
L0 ≡
and N0 ≡
1 ∂2 , 2 ∂x20
N1 ≡
∂2 , ∂x0 ∂x2
and N2 ≡
1 ∂2 ∂2 + . 2 ∂x21 ∂x0 ∂x2
Utilizing these operators, equation (12.7.9) reduces to a system of perturbation equations in the form: O(ε) : Lu1 = 0, O ε2 : L0 u2 + L1 u1 = N0 u21 , O ε3 : L0 u3 + L1 u2 + L2 u1 = N0 [2u1 u2 ] + N1 u21 , O ε4 : L0 u4 + L1 u3 + L2 u2 + L3 u1 = N0 u22 + 2u1 u3 + N1 [2u1 u2 ] + N2 u21 .
(12.7.10) (12.7.11) (12.7.12) (12.7.13)
We consider a strongly dispersive wave system, and it is necessary to take into account nonlinear modulation of the wavetrain; that is, amplitude and phase suffer from a slow variation over a localized region in space. Consequently, it is reasonable to assume that the first-order perturbation solution of the first equation (12.7.10) has the form u1 = A(x1 , x2 , . . . , xm , t1 , t2 , . . . , tm ) exp(iθ) + c.c.,
(12.7.14)
where θ = (kx0 − ωt0 ), the amplitude A is a complex function of the slow variables, and c.c. represents the complex conjugate of the preceding expression. Moreover, k and ω satisfy the dispersion relation (12.7.7). Invoking (12.7.14) in the second equation (12.7.11) of the perturbation system yields ∂D ∂A ∂D ∂A L0 u2 = −i − exp(iθ) − 2k 2 A2 exp(2iθ) + c.c., (12.7.15) ∂ω ∂t1 ∂k ∂x1 where ∂D = −2 1 + λk2 ω, ∂ω
∂D = 2 c2 − λω 2 k. ∂k
(12.7.16ab)
To determine u2 from (12.7.15), it is necessary to integrate it, so that the result produces what is called a secular (unbounded) term, that is, the term grows without
12.7 The Method of Multiple Scales
653
bound, as time t → ∞, due to the form of L0 and exp(iθ) in the first term on the right-hand side of (12.7.15). This means that such a solution for u2 will not be uniformly valid as t → ∞, no matter how small ε may be. To eliminate the nonuniformity in the perturbation expansion, the coefficient of exp(iθ) in (12.7.15) must vanish identically. This leads to the following condition for nonsecularity: ∂A ∂A + Cg (k) = 0, ∂t1 ∂x1 where Cg (k) =
Dk ∂ω =− ∂k Dω
(12.7.17)
(12.7.18)
represents the group velocity of the linearized problem. Consequently, the uniformly valid solution for u2 can be obtained from (12.7.15) in the form 2(kA)2 exp(2iθ) + F (x1 , x2 , . . . , xm , t1 , t2 , . . . , tm ) exp(iθ) + c.c. D(2ω, 2k) + G(x1 , x2 , . . . , xm , t1 , t2 , . . . , tm ), (12.7.19)
u2 = −
where F and G are complex and real functions, respectively, of higher-order scales that can be found from higher-order perturbations. Substituting u1 and u2 in equation (12.7.12) leads to another nonsecular (bounded) condition ∂D ∂A ∂D ∂A i · · − ∂ω ∂t2 ∂k ∂x2 2 1 ∂ D ∂2A ∂ 2A ∂2D ∂ 2D ∂2A − · · −2 + · 2 ∂ω 2 ∂t21 ∂k∂ω ∂x1 ∂t1 ∂k 2 ∂x21 ! 2k 2 A2 A∗ + GA + k2 − D(2ω, 2k) ∂D ∂F ∂D ∂F = 0, (12.7.20) − +i ∂ω ∂t1 ∂k ∂x1 and its complex conjugate relation. The variation of G follows from the nonsecular condition obtained from the fourth equation (12.7.13) of the perturbation system, and hence, it turns out that ∂2G ∂ 2 (AA∗ ) ∂2G − c2 2 = . 2 ∂t1 ∂x1 ∂x21
(12.7.21)
If A and G depend only on x1 and t1 through ξ = x1 − Cg t1 , that is, if they are considered in a coordinate system moving with the group velocity Cg , equation (12.7.21) can be integrated with respect to ξ and then solved to obtain −1 AA∗ + α, G(ξ, x2 , . . . , xm , t2 , t3 , . . . , tm ) = Cg2 − c2
(12.7.22)
654
12 Asymptotic Methods and Nonlinear Evolution Equations
provided that Cg = c and α = α(x2 , x3 , . . . , xm , t2 , t3 , . . . , tm ) is a slowly varying function of higher-order slow variables to be determined from the initial and boundary conditions. However, if Cg = c, the resonant interaction between short and long waves occurs, and this case can be handled separately. To obtain an equation for the amplitude A, it is necessary to find F involved in (12.7.20). Since F is a coefficient of a secular term involved in u2 and thus is an unbounded term proportional to exp(iθ), it follows that F is associated with the unbounded terms of the higher-order solutions un . It is possible to show that these unbounded (resonant) terms can be transferred to the lower-order solution u1 and the transferred quantity then can be considered as a new amplitude A. Therefore, the last term in (12.7.20) may be dropped. Finally, using (12.7.17) and (12.7.22) in the nonsecular condition (12.7.20) yields 1 ∂2A ∂A ∂A + Cg (k) 2 + Cg i ∂t2 ∂x2 2 ∂ξ 2 1 1 k 2 |A| + + A + αA = 0, (12.7.23) Dω Cg2 − c2 6λω 2 where A = A(ξ, x2 , . . . , xm , t2 , t3 , . . . , tm ), α is independent of ξ, and hence, (12.7.22) indicates that α can be treated as a constant with respect to the ξ coordinate. In a frame of reference moving with group velocity Cg and the coordinate transformations defined by 1 (x2 − Cg t2 ) = (x1 − Cg t1 ) = ε(x2 − Cg t) ε τ = t2 = εt1 = ε2 t,
ξ=
and
(12.7.24ab)
we can express equation (12.7.23) in the following canonical form of the nonlinear Schrödinger equation: 2 ∂2A 1 k ∂A 1 1 2 + Cg (k) 2 + + A + αA = 0. (12.7.25) i |A| ∂τ 2 ∂ξ Dω Cg2 − c2 6λω 2 Thus, the upshot of this perturbation analysis is that the amplitude modulation A(ξ, τ ) for a strongly dispersive system satisfies the nonlinear Schrödinger equation. We close this section by adding some relevant comments on the asymptotic expansion in terms of another small parameter δ. First, the coordinate transformations defined by (12.7.24ab) are nothing but the Gardner–Morikawa transformations that have been used in the reductive perturbation analysis. Second, u(x, t) has a more general asymptotic expansion in the form u(x, t) =
∞
δ n un (x0 , x1 , . . . , xm , t0 , t1 , . . . , tm ),
(12.7.26)
n=1
where δ is a new small parameter characterizing the smallness of terms that measure the degree of nonlinearity. Several cases arise depending on the relative importance
12.7 The Method of Multiple Scales
655
of δ and ε. The case δ = ε has already been discussed above. The case δ = ε2 corresponds to the dispersive wave systems dominated by linear propagation of envelopes. This case can be analyzed similarly to the case of δ = ε, and hence, the wave amplitude A satisfies the linear √ equation similar to (12.7.25) without the nonlinear term |A|2 A. In the case δ = ε, the nonlinear interaction dominates, and hence, the nonlinear term occurs in the first nonsecular condition. A similar analysis leads to the following equation: 2 ∂A k 1 ∂A 2 + i + Cg |A| A + αA = 0, (12.7.27) ∂t1 ∂x1 Dω 6λω 2 where A and α are functions of slow scales x1 , x2 , . . . , xm , t1 , t2 , . . . , tm . The independent variables in (12.7.27) are the first-order slow variables x1 and t1 . The slowly varying function α can be found with the aid of the same equation (12.7.21) that was obtained from the higher-order perturbation analysis. Thus, α can be treated as a function of slow variables x2 , x3 , . . . , xm , t2 , t3 , . . . , tm to be determined from the prescribed boundary and/or initial conditions. Hence, the amplitude equation (12.7.27) in the nonsecular condition remains valid for the first-order slow variables. Additional slower variables x2 , t2 , x3 , t3 , . . . in (12.7.27) can be treated as slow parameters for this order of approximation. Inoue and Matsumoto (1974) applied the method of multiple scales to a general system of equations of the form A(U )
∂U ∂U + B(U ) + C(U ) = 0, ∂t ∂x
(12.7.28)
where U is a column vector with n components u1 , u2 , . . . , un , A and B are n × n matrices, and C is a column vector. According to the method of multiple scales, we introduce a new set of independent variables as xn = εn x and tn = εn t, (12.7.29ab) where n = 0, 1, 2, . . . , m and, then, the dependent variable U is expanded in the form U = U0 +
m+1
εn Un (x0 , x1 , . . . , xm , t0 , t1 , . . . , tm ) + O εm+2 ,
(12.7.30)
n=1
where U0 is a constant solution vector which satisfies the compatibility condition C(U0 ) = 0.
(12.7.31)
Invoking the asymptotic expansion of the derivatives into multiple scale independent variables, as in (12.7.3ab), the perturbation analysis for the general system of equations can be carried out without any difficulty. However, we will not pursue the analysis further. For more details of perturbation analysis, the reader is referred to Inoue and Matsumoto (1974) and Jeffrey and Kawahara (1982).
656
12 Asymptotic Methods and Nonlinear Evolution Equations
In passing we may mention that other investigators (Leibovich and Seebass 1972) have also applied the method of multiple scales in deriving the KdV and Burgers equations. Prasad and Ravindran (1977) have developed a more general method to derive similar model equations for the propagation of nonlinear curved waves in several dimensions.
12.8 Asymptotic Expansions and Method of Multiple Scales As we have seen in this chapter, the method of asymptotic expansion is fairly direct and requires no deep knowledge of the theory of the subject. Despite the long history of the subject, the method has not been free from controversy, mainly over the interpretation of divergent series which may arise in applied mathematics. The method is very successful and is often used based on the assumption that a solution of the given equation exists as a suitable asymptotic expansion with respect to the relevant small parameter. This section deals with examples of applications of asymptotic methods to the solution of partial differential equations in the context of problems in wave propagation. We again use the Boussinesq equation (12.7.6) to illustrate this method. The dispersion relation for this equation shows that ω → 0 in the long-wavelength limit ∂ as k → 0. It is then evident from the expansions (12.7.4ab) that both operators ∂x 0 and ∂t∂0 can be removed so that L0 = L1 = N0 = N1 = 0. Consequently, the system of operators (12.7.4ab) reduces to L2 ≡
2 ∂2 2 ∂ − c , ∂t21 ∂x21
(12.8.1)
∂2 ∂2 − 2c2 , ∂t1 ∂t1 ∂x1 ∂x2 2 ∂ ∂2 ∂2 ∂2 ∂4 2 L4 ≡ 2 + 2 −c +2 −λ 2 2 , ∂t2 ∂t1 ∂t3 ∂x22 ∂x1 ∂x3 ∂t1 ∂x1 L3 ≡ 2
and N2 ≡
1 ∂2 2 ∂x21
and
N3 ≡
∂2 . ∂x1 ∂x2
(12.8.2) (12.8.3)
(12.8.4ab)
We write the asymptotic expansion for u(x, t) in the form u(x, t) =
M
δ n un (x1 , x2 , . . . , xm , t1 , t2 , . . . , tm ) + O δ M +1 .
(12.8.5)
n=1
Several possible cases arise depending on the relative importance of dispersion and nonlinearity in this problem.
12.8 Asymptotic Expansions and Method of Multiple Scales
657
Case (i): δ = ε2 . In this case, the Boussinesq equation reduces to the following coupled system: (12.8.6) O ε4 : L2 u1 = 0, 5 O ε : L3 u1 = 0, (12.8.7) 6 2 O ε : L2 u2 + L4 u1 = N2 u1 . (12.8.8) Equation (12.8.6) yields the linear wave equation 2 2 ∂ 2 ∂ u1 = 0. −c ∂t21 ∂x21
(12.8.9)
This admits a solution for u1 which is a function of ξ1 = x1 − ct1 for waves propagating to the right, or of ξ2 = x1 + ct1 for waves traveling to the left with constant velocity c. Using these results in (12.8.7) combined with terms of order O(ε5 ) yields ∂ ∂ ∂ +c (12.8.10) u1 (ξ1 , x2 , . . . , xm , t2 , t3 , . . . , tm ) = 0. ∂t2 ∂x2 ∂ξ1 This equation is satisfied if u1 depends on x2 and t2 through ξ2 = x2 − ct2 , and hence, equation (12.8.8) reduces to the form 2 2 4 ∂2 1 ∂ 2 u21 ∂ ∂2 2 ∂ 2 2 ∂ + 2c + λc . 2 −c 2 u2 − 2c 4 u1 = ∂t1 ∂x1 ∂ξ1 ∂t3 ∂ξ1 ∂x3 ∂ξ1 2 ∂ξ12 (12.8.11) Finally, we assume that u2 depends on x1 and t1 through ξ1 = x1 − ct1 so that equation (12.8.11) with u1 = φ takes the form ∂φ ∂ 3φ ∂φ ∂φ 1 1 φ +c + (λc) 3 + = 0, (12.8.12) ∂t3 ∂x3 2 ∂ξ1 2c ∂ξ1 in which integration is carried out with respect to ξ1 . In a frame of reference moving with the phase velocity c of long waves, that is, by setting ξ3 = x3 −ct3 , τ = t3 and, then, replacing ξ1 by ξ, equation (12.8.12) reduces to the canonical form of the KdV equation in terms of the independent variables ξ and τ ∂φ ∂φ 1 1 ∂3φ φ + (λc) 3 = 0. + (12.8.13) ∂τ 2c ∂ξ 2 ∂ξ This reveals the fact that, when deriving the KdV equation, the frame of reference moves with the phase velocity but not with the group velocity of the wave. Case (ii): δ = ε3 . In this case, the asymptotic expansion enables us to derive an equation similar to (12.8.12) without the nonlinear term.
658
12 Asymptotic Methods and Nonlinear Evolution Equations
Case (iii): δ = ε. In this case, we obtain a nonlinear equation without a dispersive term in the form ∂φ ∂φ 1 ∂φ φ = 0. (12.8.14) +c + ∂t2 ∂x2 2c ∂ξ Obviously, the independent variables are the lower-order scales x2 and t2 in equation (12.8.14). Example 12.8.1 (The Korteweg–de Vries (KdV) Equation). We consider the nonlinear partial differential equation (Johnson 1997) utt − uxx = ε(u2 + uxx )xx ,
x ∈ R, t ≥ 0,
(12.8.15)
where ε is a small parameter that represents the characteristics of both small amplitude and long waves. We seek a solution of (12.8.15) in terms of the small parameter ε as u(x, t; ε) =
∞
εn un (x, t)
as ε → 0,
(12.8.16)
n=0
where both x and t are O(1). We assume that equation (12.8.15) is to be solved with the appropriate initial condition at t = 0. We substitute (12.8.16) in (12.8.15), collect together like powers of ε, and then set equal to zero for each coefficient of εn . Consequently, (12.8.16) is a solution of (12.8.15) provided u0tt − u0xx = 0, u1tt − u1xx = u20 + u0xx xx .
(12.8.17) (12.8.18)
Clearly, the d’Alembert solution of (12.8.17) is given by u0 (x, t) = f (x − t) + g(x + t),
(12.8.19)
where f and g are arbitrary functions of their arguments. For simplicity, we assume that the initial data is such as to generate only the wave traveling to the positive x-direction so that u(x, 0; ε) = f (x),
ut (x, 0; ε) = −f (x).
With u0 = f (x − t) as the solution, equation (12.8.18) becomes u1tt − u1xx = f 2 + f ,
(12.8.20)
(12.8.21)
where the prime denotes the derivative with respect to (x − t). We next introduce the characteristic variables ξ, η for (12.8.21) ξ = x − t,
η = x + t,
(12.8.22)
12.8 Asymptotic Expansions and Method of Multiple Scales
so that
−4u1ξη = f 2 + f ,
659
(12.8.23)
and hence,
1 u1 (ξ, η) = − η f 2 + f + A(ξ) + B(η), (12.8.24) 4 where f = f (ξ), and the arbitrary functions A and B are determined from the initial data (12.8.20) with the following requirements for u1 (x, t): u1 (x, 0) = 0 = u1t (x, 0).
(12.8.25)
Consequently, we have u1 (ξ, η) =
1 (ξ − η) f 2 (ξ) + f (ξ) − f 2 (ξ) − f (ξ) + f 2 (η) + f (η) , 4
and therefore, u1 (x, t) =
1 1 F (x + t) − F (x − t) − tF (x − t), 4 2
(12.8.26)
where F = f 2 + f . Thus, the asymptotic expansion is u(x, t; ε) = f (x − t) −
ε 2tF (x − t) + F (x − t) − F (x + t) . 4
(12.8.27)
If f has compact support, or at least for f (x) → 0 sufficiently rapidly as x → ±∞, the asymptotic result (12.8.27) is not uniformly valid for εt = O(1). We now investigate the solution of (12.8.15) using the transformations ξ = x − t = O(1),
τ = εt = O(1)
as ε → 0.
(12.8.28)
It is noted that the asymptotic expansion (12.8.27) would be nonuniform (breaking down) for any values of ξ where the first, second, or third derivative of f (ξ) is not defined. The solution (12.8.27) is also nonuniform for any t, no matter how well behaved f (ξ) may be. Furthermore, the solution is of no physical interest for a constant value of f . In wave propagation problems, the region where a large time (or distance) variable τ is involved is called the far-field region, and the region where t = O(1) is then referred to as the near-field region. We also observe that, for ξ = x − t = O(1), t = O(ε−1 ) implies that x = O(ε−1 ). We next apply the transformation (12.8.28) to equation (12.8.15) and use the fact that ∂ ∂ξ ∂ ∂τ ∂ ∂ = · + · = , ∂x ∂ξ ∂x ∂τ ∂x ∂ξ ∂ ∂ξ ∂ ∂τ ∂ ∂ ∂ = · + · =ε − , ∂t ∂ξ ∂t ∂τ ∂t ∂τ ∂ξ
660
12 Asymptotic Methods and Nonlinear Evolution Equations
so that the equation for u(x, t; ε) = U (ξ, τ ; ε) reduces to the form εUτ τ − 2Uτ ξ = U 2 + Uξξ ξξ .
(12.8.29)
We seek an asymptotic solution of this equation in the form U (ξ, τ ; ε) =
∞
εn Un (ξ, τ )
as ε → 0,
(12.8.30)
n=0
where ξ = O(1) and τ = O(1). Thus, U0 (ξ, τ ) satisfies the equation 2U0τ ξ + U02 + U0ξξ ξξ = 0,
(12.8.31)
or equivalently, 2(U0τ + U0 U0ξ ) + U0ξξξ = 0,
(12.8.32)
where decay conditions are imposed as ξ → ±∞. This is one form of the Korteweg–de Vries equation. The solution of this equation satisfying the matching condition U0 → f (ξ)
as τ → 0
corresponds to the initial-value problem for equation (12.8.32). Such a solution exists, provided f (ξ) decays sufficiently rapidly as ξ → ±∞. The inverse scattering method was used to obtain the solution in Chapter 9. The solution for U0 (ξ, τ ) represents a one-term uniformly valid asymptotic expansion for τ ≥ 0 and τ = O(1) as ε → 0. Finally, the next term in the expansion (12.8.30) satisfies the equation 2U1τ ξ + 2(U0 U1 )ξξ + U1ξξξξ = U0τ τ , or equivalently, 2U1τ + 2(U0 U1 )ξ + U1ξξξ = − U02 + U0ξξ τ ,
(12.8.33)
where equation (12.8.32) was used for U0τ and the decay conditions have been imposed as ξ → ±∞. The mathematical analysis hereafter is not particularly simple. However, the solution for U1 is found by writing U1 = U0ξ V (ξ, τ ), which can be investigated to check whether the asymptotic expansion (12.8.30) is valid as τ → ∞. This involves a lengthy and somewhat complicated discussion if the general term Un is to be included. We will not pursue this problem further. We conclude by making a final statement that the far-field asymptotic expansion of the solution for problems of this type is usually uniformly valid, provided f (x) is sufficiently smooth and decays to zero sufficiently rapidly at infinity. Finally, we describe the method of multiple scales to solve another example which is typical of some problems in water waves.
12.8 Asymptotic Expansions and Method of Multiple Scales
661
Example 12.8.2 (The Nonlinear Schrödinger equation). We consider the nonlinear partial differential equation (Johnson 1997) utt − c2 uxx − u + ε(uux )x = 0,
(12.8.34)
where ε is a small parameter. For ε = 0 this equation admits traveling wave solutions in the form u(x, t) = A exp i(ωt − kx) + c.c., (12.8.35) where c.c. represents the complex conjugate, and A is the complex amplitude of the wave. Substituting (12.8.35) in (12.8.34) with ε = 0 gives the dispersion relation ω2 = c2 k 2 − 1,
(12.8.36)
which admits a real solution for ω if |ck| > 1. We assume ck > 1 so that there are two possible phase velocities of the wave given by " ω (12.8.37) cp = = ± c2 − k −2 . k For a given k and a phase velocity cp , there exists a harmonic wave solution of equation (12.8.34) which evolves slowly on suitable scales. It is convenient to introduce slow variables ζ = ε(x − cg t), τ = ε2 t, (12.8.38) where the group velocity cg is, in general, not equal to the phase velocity cp and is known at this point. In addition, introducing a new variable ξ = x − cp t, and using the following identities: ∂ ∂ ∂ = +ε , ∂x ∂ξ ∂ζ
∂ ∂ ∂ ∂ = −cp − εcg + ε2 , ∂t ∂ξ ∂ζ ∂τ
in the original equation (12.8.34) with c = 1 for u(x, t; ε) = U (ξ, ζ, τ ; ε) reduces to the form 2 cp − 1 Uξξ − U + 2ε(cp cg − 1)Uξζ + ε2 c2g − 1 Uζζ − 2cp Uξτ (12.8.39) + ε(U Uξ )ξ + ε2 (U Uξ )ζ + (U Uζ )ξ = O ε3 , where only terms O(ε2 ) are retained in the above equation. Thus, the function U (x, t; ε) is now treated as a function of the new variables ξ, ζ, τ . In view of the scales O(1), O(ε−1 ), and O(ε−2 ), the method adopted here is called the method of multiple scales. We now seek a solution in the form of the asymptotic expansion U (ξ, ζ, τ ; ε) =
∞ n=0
where ξ, ζ, and τ are all O(1).
εn Un (ξ, ζ, τ )
as ε → 0,
(12.8.40)
662
12 Asymptotic Methods and Nonlinear Evolution Equations
Consequently, U0 satisfies the linear equation 2 cp − 1 U0ξξ − U0 = 0.
(12.8.41)
We seek a solution in the form U0 (ξ, ζ, τ ) = A01 (ζ, τ )eikξ + c.c. with c2p
1 = 1− 2 , k
k > 1,
(12.8.42)
(12.8.43)
where the first subscript in A01 represents the term ε0 , and the second term is related to the choice E m = exp(imkξ), m = 1, 2, . . . . At the next order ε1 , we find the nonlinear equation 2 cp − 1 U1ξξ − U1 = 2(1 − cp cg )U0ξζ + (U0 U0ξ )ξ = (1 − cp cg )(ikA01 E + c.c.) − 2k2 A201 E 2 + c.c. , (12.8.44) and U1 is a harmonic function only if cp cg −1 = 0, which determines cg . If we do not make this choice for cg , then U1 includes a particular integral proportional to (ξE) which would lead to the nonuniform asymptotic expansion (12.8.40) as |ξ| → ∞. Terms like (ξE) are called secular, while a uniform solution in ξ is possible only if harmonic terms in ξ are allowed in the solution for U1 . The amplitude A01 of the wave moves with the group velocity cg which satisfies the condition cp cg = 1, which can also be verified from the definition of the group velocity cg =
d dω = (kcp ) dk dk
with (12.8.43).
Thus, the solution for U1 (ξ, ζ, τ ) can be written as U0 (ξ, ζ, τ ) = A11 (ζ, τ )E +
2k 2 A201 E 2 + c.c. − c2p ) − 1
4k 2 (1
2 = A11 (ζ, τ )E + k 2 A201 E 2 + c.c., 3
(12.8.45)
where A11 is the amplitude of the fundamental wave E = exp(ikξ). It is noted that U1 includes a higher harmonic E 2 and its complex conjugate E −2 . With the condition cp cg = 1, we obtain the equation for U2 (ξ, ζ, τ ) as 2 cp − 1 U2ξξ − U2 = 1 − c2g U0ζζ + 2cp U0ξτ − U02 ξζ − (U0 U1 )ξξ . (12.8.46) We impose the condition that the solution for U2 must contain terms harmonic in ξ so that any terms in E 1 involved in the forcing terms in (12.8.46) must vanish. Such terms can arise only from the equation
12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations
663
1 − c2g U0ζζ + 2cp U0ξτ − (U0 U1 )ξξ = 1 − c2g [A01ζζ E + c.c.] + 2cp (ikE)A01τ + c.c. ∂2 − 2 A01 E + A01 E −1 ∂ξ ! 2 2 2 2 2 −2 −1 + A11 E + A11 E + , (12.8.47) k A01 E + A01 E 3
where the bar denotes the complex conjugate. The coefficient of E involved in this expression must vanish (and, of course, its conjugate for E −1 ) to satisfy the condition imposed on U2 so that 2 (2ikcp )A01τ + 1 − c2g A01ζζ + k 4 A01 |A01 |2 = 0, 3
(12.8.48)
where all the other terms generate higher harmonics in U2 , which is acceptable for uniformity in the asymptotic solution as |ξ| → ∞. This equation (12.8.48) is well known as the nonlinear Schrödinger (NLS) equation, which describes the nonlinear evolution of the wave amplitude.
12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations This section deals with two derivations that lead to a description of the evolution of wavepackets for surface gravity waves on water of finite depth. We use an asymptotic method to derive first the (1 + 1)-dimensional nonlinear Schrödinger equation and then (2 + 1)-dimensional Davey–Stewartson equations. Example 12.9.1 (Derivation of the NLS Equation from the Laplace Equation for the Velocity Potential). We recall the governing equation (9.3.3)–(9.3.6) for nonlinear water waves with no y-dependence in the form φzz + δφxx = 0, 1 φt + η + ε(φ2x + δ −1 φ2z ) = 0 on z = 1 + εη, 2 φz = δ(ηt + εφx ηx ) on z = 1 + εη, φz = 0 on z = 0,
(12.9.1) (12.9.2) (12.9.3) (12.9.4)
where parameters δ and ε are given by (9.3.2ab). We next introduce the new variables ξ = x − cp t,
ζ = ε(x − cg t),
where cp and cg are to be determined.
τ = ε2 t,
(12.9.5)
664
12 Asymptotic Methods and Nonlinear Evolution Equations
In view of the transformations (12.9.5), equations (12.9.1)–(12.9.4) reduce to (12.9.6) φzz + δ φξξ + 2εφξζ + ε2 φζζ = 0, ε ε2 φτ + η − εcg φζ + cp φξ ) + δ −1 φ2z + (φξ + εφζ )2 = 0 2 on z = 1 + εη, (12.9.7) 2 φz = δ ε ητ − (cp ηξ + εcg ηζ ) + ε(φξ + εφζ )(ηξ + εηζ ) on z = 1 + εη, (12.9.8) φz = 0 on z = 0.
(12.9.9)
We seek an asymptotic solution of this system in the form ∞
φ=
n
ε φn (ξ, ζ, τ, z),
n=0
η=
∞
εn ηn (ξ, ζ, τ )
as ε → 0,
(12.9.10)
n=0
which are periodic functions in ξ. In the leading order O(1), the problem is given by φ0zz + δφ0ξξ = 0, η0 − cp φ0ξ = 0, φ0z = −δcp η0ξ φ0z = 0 on z = 0.
on z = 1,
(12.9.11) (12.9.12) (12.9.13)
We seek the solution of this leading order system as φ0 = f0 (ζ, τ ) + Φ0 (z, ζ, τ )E + c.c.,
η0 = A0 (ζ, τ )E + c.c.,
(12.9.14)
where E = exp(ikξ) and c.c. denotes the complex conjugate of the terms in E. Clearly, the above solution describes a harmonic wave of wavenumber k propagating with the velocity cp . Evidently, the Laplace equation (12.9.11) with (12.9.14) becomes Φ0zz = δk 2 Φ0 .
(12.9.15)
The solution of (12.9.15) with the bottom boundary condition (12.9.13) is √ (12.9.16) Φ0 = F0 (ζ, τ ) cosh δkz , where the function F0 = F0 (ζ, τ ) is to be determined. The two surface boundary conditions (12.9.12) yield √ √ √ δkF0 sinh δk = −iδkcp A0 , ikcp F0 cosh δk = A0 ,
(12.9.17)
which give c2p
√ tanh δk , = √ δk
√ iA0 sech δk . F0 = − kcp
(12.9.18ab)
12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations
665
Thus, solution (12.9.16) becomes
√ √ cosh( δkz) √ , Φ0 = −i δcp A0 sinh( δk)
(12.9.19)
where the wave amplitude A0 = A0 (ζ, τ ) is to be determined. At the next order O(ε), we proceed to collect all terms and use the surface boundary conditions on z = 1 to obtain (12.9.20)
φ1zz + δ(φ1ξξ + 2φ0ξζ ) = 0,
1 η1 − cg φ0ζ − cp (φ1ξ + φ0ξz ) + δ −1 φ20z + φ20ξ = 0 2 on z = 1, φ1z + η0 φ0zz = δ(φ0ξ η0ξ − cg η0ζ − cp η1ξ ) on z = 1, φ1z = 0
on z = 0.
(12.9.21) (12.9.22) (12.9.23)
It is evident that the nonlinearity of the surface boundary conditions generates terms E 0 , E 2 , and E −2 from equations (12.9.20)–(12.9.23). The nonlinear interactions produce higher harmonics E 2 , E −2 and their complex conjugates in addition to the fundamental harmonics E 1 with E −1 introduced by (12.9.14). We seek a periodic solution in ξ and remove all secular terms that contribute to the nonperiodic terms to the solution. In order to carry out this program further, we employ the asymptotic procedure similar to that used in Examples 12.8.1 and 12.8.2. Thus, we write φm =
m+1
Φmn E n + c.c.,
n=0
ηm =
m+1
Amn E n + c.c.,
(12.9.24)
n=0
where m = 1, 2, 3, . . . , Φmn (z, ζ, τ ) and Amn (ζ, τ ) are yet to be determined, and c.c. refers only to terms E n , n = 1, 2, 3, . . . . The terms corresponding to n = 0 are not harmonic (or oscillatory) solutions, but they are periodic. At each higher order in ε, higher-order harmonics are produced so that E 2 appears first at O(ε), E 3 first at O(ε2 ), and so on. The Laplace equation (12.9.20) for φ1 yields Φ10zz = 0,
Φ12zz − 4δk 2 Φ12 = 0
(12.9.25)
and Φ11zz − δk2 Φ11 + 2iδkΦ0ζ = 0.
(12.9.26)
These equations have solutions satisfying the bottom boundary condition (12.9.23) as √ Φ10 = F10 (ζ, τ ), Φ12 = F12 (ζ, τ ) cosh 2 δkz , (12.9.27) √ √ √ (12.9.28) Φ11 = F11 (ζ, τ ) cosh δkz − i δF0ζ z sinh δkz , where F1n (ζ, τ ) are arbitrary functions of ζ and τ . These solutions are then utilized in the two free surface boundary conditions (12.9.21), (12.9.22) to obtain six equations arising from coefficients of E 0 , E 1 and E 2 . Equation (12.9.21) gives
666
12 Asymptotic Methods and Nonlinear Evolution Equations
√ √ E 0 : − cg f0ζ + i δk 2 cp (A0 F 0 + A0 F0 ) sinh δk √ √ (12.9.29) + A10 + k 2 F0 F 0 sinh2 δk + cosh2 δk = 0, √ √ √ √ E 1 : − cg F0ζ cosh δk − ikcp F11 cosh δk − i δF0ζ sinh δk (12.9.30) + A11 = 0, √ √ √ E 2 : − ikcp 2F12 cosh 2 δk + δkA0 F0 sinh δk 1 + A12 − k2 F02 = 0, (12.9.31) 2 and similarly, equation (12.9.22) leads to √ E 0 : (A0 F 0 + A0 F0 )δk2 cosh δk
√ = δk2 (A0 F 0 + A0 F0 ) cosh δk, (12.9.32) √ √ √ √ √ √ 1 E : δkF11 sinh δk − i δF0ζ sinh δk + δk cosh δk (12.9.33) = −δ(cg A0ζ + ikcp A11 ), √ √ √ 2 2 E : 2 δkF12 sinh 2 δk + δk A0 F0 cosh δk √ (12.9.34) = −δ 2ikcp A12 + k 2 A0 F0 cosh δk , where the bar denotes the complex conjugate. It is noted that (12.9.32) is identically satisfied and that with result (12.9.18ab) and for Φ0 used in equation (12.9.29), we find √ 2 δk √ A0 A0 + cg f0ζ . A10 = − (12.9.35) sinh 2 δk Equation (12.9.30) leads us directly to obtain √ √ √ √ A11 = cg F0ζ cosh δk + ikcp F11 cosh δk − i δF0ζ sinh δk .
(12.9.36)
If this result (12.9.36) is utilized in equation (12.9.33), we see that F11 cancels identically. If we next use the result for cp in (12.9.18a) first and then F0 from (12.9.18b) in (12.9.33), it turns out that A0ζ also cancels and the result becomes cg =
√ √ 1 cp 1 + 2 δk cosech 2 δk . 2
(12.9.37)
This is the group velocity for water waves. Finally, we use (12.9.18ab) to solve (12.9.31) and (12.9.34) for F12 and A12 so that δkcp A20 3i √ F12 = − , 4 sinh4 ( δk) √ √ (12.9.38) √ 2 δk cosh( δk) 2 √ 2 cosh δk + 1 A0 , A12 = 2 sinh3 ( δk) where A0 = A0 (ζ, τ ) is to be determined.
12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations
667
In the next order O(ε2 ), equation (12.9.6) and the expansion of the free surface boundary conditions (12.9.7), (12.9.8) give (12.9.39) φ2zz + δ(φ2ξξ + 2φ1ξζ + φ2ζζ ) = 0, 1 φ0τ − cg η0 φ0ζz − cp φ1ζ + φ2ξ + η0 φ1ξz + η1 φ0ξz + η02 φ0ξzz + η2 2 + δ −1 (η0 φ0zz + φ1z )φ0z + (η0 φ0ξz + φ1ξ + φ0ζ )φ0ξ = 0 on z = 1, (12.9.40) 1 2 φ2z + η0 φ1zz + η0 φ0zzz + η1 φ0zz − δ −1 η0τ − (cg η1ζ + cp η2ξ ) 2 + (η1ξ + η0ζ )φ0ξ + (η0 φ0ξz + φ1ξ + φ0ζ )η0ξ = 0 on z = 1, (12.9.41) and φ2z = 0
on z = 0.
(12.9.42)
It is important to point out that the periodic solution adopted in (12.9.24) is now utilized for m = 2 so that the higher harmonics E 3 now appears for the first time in this analysis. Consequently, equation (12.9.39) then yields Φ20zz + δf0ζζ = 0,
Φ21zz − δk 2 Φ21 + δ(2ikΦ11ζ + Φ0ζζ ) = 0,
(12.9.43)
and so on. The solution for Φ21 (z, ζ, τ ) satisfying the bottom boundary condition (12.9.42) then becomes √ √ √ 1 F0ζζ z sinh δkz Φ21 = F21 cosh δkz − δ iF11ζ + 2k √ √ z 1 2 + δF0ζζ √ sinh δkz − z cosh δkz . (12.9.44) 2 k δ The free surface boundary condition (12.9.41) reduces to, for terms E 1 , 1 Φ21z + A0 Φ12z + A20 Φ0zzz + 2A0 A0 Φ0zzz + A10 Φ0zz + A12 Φ0zz 2 = δ A0τ − (cg A11ζ + ikcp A21 ) + 2k2 A12 Φ0zz − k2 A0 A0 Φ0z − A0 Φ0z (12.9.45) + k2 A0 (A0 Φ0z + 2Φ12 ) on z = 1. The free surface boundary condition (12.9.40) on z = 1 gives, for terms E 1 , Φ0τ − cg Φ11ζ − ik cp Φ21 + 2cp A0 Φ12z + cp (A10 Φ0z − A12 Φ1z ) 1 − ikcp 2A0 A0 Φ0zz − A20 Φ0zz + A21 2 + δ −1 A0 Φ0zz + A0 Φ0zz Φ0z + (A0 Φ0zz + Φ12z )Φ0z − k 2 A0 Φ0z − A0 Φ0z Φ0 + k2 (A0 Φ0z + 2Φ12 )Φ0 = 0. (12.9.46)
668
12 Asymptotic Methods and Nonlinear Evolution Equations
Clearly, the equation for A0 arises from the terms which occur after elimination of secular terms at E 1 . The procedure is relatively simple to describe, but rather tedious to discuss. However, we can eliminate A12 from (12.9.45) and (12.9.46), and then introduce the functions obtained before including Φ21 from (12.9.44). Simplifying the result so that F21 cancels identically due to the definition of cp , and A11 (or F11 ) also cancels after substitution of cg from (12.9.37), the final equation for A0 (ζ, τ ) can be obtained in the form −2ikcp A0τ + αA0ζζ + β|A0 |2 A0 = 0,
(12.9.47)
where α = c2g − (1 − s tanh s)sech2 s = −(kcp ) and
d2 (kcp ), dk 2
ω(k) = kcp , (12.9.48)
1 1 + 9 coth2 s − 13 sech2 s − 2 tanh4 s β= 2 2 2 2 −1 2cp + cg sech s , − 1 − cg
k2 c−2 p
(12.9.49)
√ with s = δk. Equation (12.9.47) is well known as the nonlinear Schrödinger (NLS) equation for the evolution of nonlinear water waves, and it is one of the completely integrable √ types of equations. It is easy to check that α > 0 for all s = k δ, but β changes its √ sign from positive to negative as s decreases its value across s = k δ ≈ 1.363. It is important to point out that the nature of the solutions of (12.9.47) depend on the signs of α and β. Indeed, the condition βα−1 > 0 (αβ > 0) ensures the existence of a solitary wave solution of the NLS equation (12.9.47), and the soliton decays to zero at infinity. There exists some experimental and numerical evidence in support of the solitary wave solution. On the other hand, when β < 0 (αβ < 0), α no such solitary wave solution exists. In other words, the solution grows exponentially as τ → ∞, that is, the solution becomes unstable. Example 12.9.2 (Derivation of the Davey–Stewartson (DS) Equations). Example 12.9.1 deals with the NLS equation which describes the modulation of the amplitude of the nonlinear water waves propagating only in one direction. We closely follow Davey and Stewartson (1974) to consider the two-dimensional problem of the propagation of a plane wave. This two-dimensional problem incorporates slow (or weak) dependence on both x- and y-coordinates with the fast oscillations only in the x-direction. We present the evolution of wavepackets that will propagate in the x-direction with slowly varying structure in both x- and y-directions. However, the group velocity is still associated with the wave propagation in the x-direction.
12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations
669
We recall governing equations (9.3.3)–(9.3.6) and then introduce the variables y ∗ = εy,
ζ = ε(x − cg t),
ξ = x − cp t,
τ = ε2 t.
(12.9.50)
Consequently, equations (9.3.3)–(9.3.6) become, dropping the asterisk in y∗ , (12.9.51) φzz + δ φξξ + 2εφξζ + ε(φζζ + φyy ) = 0, 1 1 ε2 φτ − εcg φζ − cp φξ + η + ε φ2z + (φξ + εφζ )2 + εφ2y = 0 2 δ on z = 1 + εη, (12.9.52) φz = δ ε2 ητ − (εcg ηζ + cp ηξ ) + ε(φξ + εφζ )(ηξ + εηζ ) + ε3 φy ηy on z = 1 + εη, (12.9.53) and φz = 0
on z = 0.
(12.9.54)
Proceeding to collect all terms not larger than O(ε2 ), the only contribution from the y-dependence will come from the term φyy in the Laplace equation (12.9.51). All other terms involving derivatives of y produce new nonlinear interactions that will arise first at O(ε3 ). We can use the calculation similar to that already presented in Example 12.9.1 for the nonlinear Schrödinger equation. Thus, we avoid the details here and present only major steps to derive a pair of Davey–Stewartson equations. We seek asymptotic solutions in the form m+1 ∞ φ = f0 (ζ, y, τ ) + εm Φmn (z, ζ, y, τ )E n + c.c. , (12.9.55) η=
∞ m=0
ε
m
m+1
m=0
n=0
n
Amn (ζ, y, τ )E + c.c. ,
(12.9.56)
n=0
where E = exp(ikξ) and A00 = 0 so that the first-order approximation to the surface gravity waves is purely harmonic. We expect results already obtained for all the terms at O(ε0 ) and O(ε). The main differences will appear at O(ε2 ), and the problem at ε2 E 0 yields an equation for f0 ,
√ 1 1 − c2g f0ζζ + f0yy = − 2 2cp + cg sech2 δk |A0 |2 ζ , cp
(12.9.57)
where A0 = A01 is given. The free surface boundary conditions for the terms ε2 E give −2ikcp A0τ + αA0ζζ − cp cg A0yy 1 1 + 9 coth2 s − 13 sech2 s − 2 tanh4 s A0 |A0 |2 + k2 c−2 2 p + k 2 2cp + cg sech2 s A0 f0ζ = 0, (12.9.58) √ where s = δk.
670
12 Asymptotic Methods and Nonlinear Evolution Equations
These two equations (12.9.57), (12.9.58) are called the Davey–Stewartson (DS) equations for the modulation of two-dimensional harmonic waves. Using α, β given by (12.9.48) and (12.9.49) and introducing a new parameter γ by (12.9.59) γ = 2cp + cg sech2 s, the Davey–Stewartson equations can be written in a compact form γ 1 − c2g f0ζζ + f0yy = − 2 |A0 |2 ζ , cp − 2ikcp A0τ + αA0ζζ − cp cg A0yy γ 2 k2 A0 |A0 |2 + γk2 A0 f0ζ = 0. + β+ 2 cp (1 − c2g )
(12.9.60)
(12.9.61)
It is noted here that γ > 0 and cp cg > 0. For no y-dependence, equation (12.9.60) becomes after integration γ 2 (12.9.62) 1 − cg f0ζ = − 2 |A0 |2 . cp This equation (12.9.62) describes the mean drift generated by the nonlinear interaction of the wave motion. This is usually called the Stokes drift. In the absence of y-dependence with the assumption that f0ζ = 0, and hence, A0 ≡ 0, equation (12.9.61) reduces to the NLS equation −2ikcp A0τ + αA0ζζ + β|A0 |2 A0 = 0.
(12.9.63)
This is identical with the NLS equation (12.9.47). It has been an interesting exercise to present several conservation laws for the nonlinear Schrödinger equation in Section 10.5. However, the derivation of the conservation laws for the Davey–Stewartson equations is not easy because these equation are coupled and depend on the three variables τ , ζ, and y. In order to derive a few conservation laws, it is convenient to rewrite the DS equations (12.9.60), (12.9.61) by replacing τ by t, ζ by x, f0 by f , and A0 by A so that they read as afxx + fyy − b |A|2 x = 0, (12.9.64) −icAt + dAxx − eAyy + hA|A|2 + Afx = 0,
(12.9.65)
where a, b, c, d, e, and h are real constants and f is a real function. Equation (12.9.64) is already in the conservation form ∂ ∂ afx − b|A|2 + (fy ) = 0, ∂x ∂y which gives ∂ ∂x
∞ −∞
afx − b|A|
2
(12.9.66)
dy + [fy ]∞ −∞ = 0.
(12.9.67)
12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations
671
Under the decay conditions as |x| → ∞, this equation takes the form ∞ ∞ ∂ f dy = b |A|2 dy = g(t), (12.9.68) a ∂x −∞ −∞ where g(t) is an arbitrary function of t. If, for some appropriate value of x, the lefthand side of (12.9.68) vanishes, then g(t) = 0 for all t. Consequently, (12.9.68) gives ∞
f dy
a
∞
∞
−∞
−∞
=b
−∞
|A|2 dx dy = const.
This leads to another conservation law ∞ f dy = const.
(12.9.69)
(12.9.70)
−∞
We next take the complex conjugate of (12.9.65) to obtain icAt + dAxx − eAyy + hA|A|2 + Afx = 0.
(12.9.71)
We next multiply (12.9.71) by A and (12.9.65) by A and then subtract the latter result from the former result to find ic
∂ ∂ ∂ AAy − AAy = 0. AA + d AAx − AAx + e ∂x ∂x ∂y
This leads to the following results in the conservation form: ∞ ∞ ∂ ∂ 2 ic |A| dx + e AAy − AAy dx = 0 ∂t −∞ ∂y −∞ and ic
∂ ∂t
∞ d AAx − AAx dy = 0, |A|2 dy + d dx −∞ −∞
(12.9.72)
(12.9.73)
∞
(12.9.74)
provided decay conditions are satisfied as |x| → ∞ for fixed y, and as |y| → ∞ for fixed x. This means that waves at infinity are not parallel to either the x- or ydirections. Moreover, if decay conditions hold for |x| → ∞ and |y| → ∞, that is, the solution decays sufficiently rapidly as (x2 + y 2 ) → ∞, it turns out that ∞ ∞ |A|2 dx dy = const. (12.9.75) −∞
−∞
This represents a conserved quantity that arises for a certain class of solutions. This is usually referred to as the law of conservation of mass. Finally, we close this section by representing one of the most direct applications of the NLS and DS equations. This application deals with the stability of the Stokes waves. We recall the NLS equation (12.9.47) and rewrite it in the form
672
12 Asymptotic Methods and Nonlinear Evolution Equations
−2ikcp A0τ + αA0ζζ + βA0 |A0 |2 = 0, where α and β are given by (12.9.48) and (12.9.49). We seek a nonlinear plane wave solution of (12.9.76) in the form A0 (ζ, τ ) = A exp i(mζ − στ ) ,
(12.9.76)
(12.9.77)
where the amplitude A is a complex constant and the wavenumber m is real. This solution exists provided σ satisfies the dispersion relation 2kcp σ = β|A|2 − αm2 ,
(12.9.78)
where cp , α, and β are functions of the wavenumber k(> 0) of the primary wave described by (12.9.14) and m(> 0) is the wavenumber of the modulation. Thus, the primary wave solution becomes (12.9.79) η0 = A exp i(kx − ωt + mζ − στ ) + c.c., where τ = ε2 t and σ is given by (12.9.78) with m = 0. The solution (12.9.79) represents the Stokes wave of constant amplitude A and wavenumber k with the dispersion relation 2kcp σ = β|A|2 . Thus, the Stokes wave assumes the form η0 (x, t) = A exp i kx − ω + ε2 σ t .
(12.9.80)
(12.9.81)
We next employ the NLS equation to examine the stability of the Stokes wave. The NLS equation describes the modulation of the amplitude of the harmonic wave represented by exp(ikx). We seek a solution which is a small perturbation of the nonlinear plane wave solution (12.9.81) so that (12.9.82) A0 = A(1 + μa) exp i(−στ + μθ) , where a = a(ζ, τ ) and θ = θ(ζ, τ ) are real perturbation functions and μ is a parameter with the dispersion relation (12.9.80). Substituting (12.9.82) into the original NLS equation (12.9.76) gives, dropping the exponential term in (12.9.82), −2ikcp μaτ + i(1 + μa)(μθτ − σ) + α μaζζ + 2iμ2 aζ θζ + μ(1 + μa) iθζζ − μθζ2 + β(1 + μa)3 |A|2 = 0.
(12.9.83)
The leading terms O(1) as μ → 0 cancel due to (12.9.80), and hence, the main perturbation terms O(μ) are given by −2ikcp aτ + i(θτ − σ) + α(aζζ + iθζζ ) + 3β|A|2 a = 0. (12.9.84)
12.9 Derivation of the NLS Equation and Davey–Stewartson Evolution Equations
673
In view of the fact that a and θ are real functions, we use (12.9.80) to eliminate σ from (12.9.84) so that 2kcp θτ + αaζζ + 2β|A|2 a = 0, −2kcp aτ + αθζζ = 0.
(12.9.85) (12.9.86)
This is a pair of linear partial differential equations with constant coefficients. We seek a solution of the form a0 a = exp i(κζ − Ωτ ) + c.c., (12.9.87) θ0 θ where a0 , θ0 , κ(> 0), and Ω are constants. This solution exists, provided the dispersion relation for Ω β 2 2 2 2 2 (2kcp Ω) = α κ κ − 2 |A| (12.9.88) α is satisfied. Or equivalently, 12 β |A|2 . (2kcp Ω) = (ακ) κ2 − 2 α
(12.9.89)
Some new and striking conclusions can be drawn from (12.9.89). If βα−1 > 0 (αβ > 0), then Ω is imaginary for some wavenumber κ, provided # √ β 0 < κ < 2|A| . (12.9.90) α In this case, the amplitude modulation grows exponentially as τ → ∞. This means that the Stokes wave is definitely unstable for a range of wavenumbers κ given by (12.9.90). On the other hand, if βα−1 < 0 (αβ < 0), Ω is real for all values of the wavenumber κ. This implies that the amplitude modulation persists, but does not grow. In other words, the Stokes wave is neutrally stable for all values of κ. Another most simple and direct interpretation of the instability phenomenon of the Stokes wave can be given by writing the leading order fundamental wave mode at t = 0 as (12.9.91) η0 = A(1 + μa) exp i(kx + μθ) + c.c., which is, by using (12.9.87) with ζ = ε(x − cg t) at t = 0, = A exp i(kx + μθ) + Aμa0 exp i(k + εκ)x + iμθ + c.c.
(12.9.92)
as μ → 0 for fixed x and ε. Evidently, the perturbation to the fundamental mode has the term μa0 which has a wavelength k + εκ that is close to k. This means that a perturbation with a wavenumber close to that of the fundamental mode will produce an unstable solution whenever βα−1 > 0. Thus, it is impossible to generate stable waves with a precisely
674
12 Asymptotic Methods and Nonlinear Evolution Equations
fixed wavenumber both in nature and in the laboratory. In other words, a wave with a small deviation from a fixed wavenumber k would always occur and lead to the instability phenomenon. This kind of instability is associated with a small change in the wavenumber of the fundamental mode and is now well known as the Benjamin– Feir (1967) side-band instability. It is often observed in nature that a set of plane waves gradually breaks down along the wave fronts into a series of wave groups.
12.10 Exercises 1. The simplest equation for compressible fluid flows is a 2 × 2 matrix of the continuity and momentum equations in two independent variables x and t, that is, or equivalently,
Ut + Fx = S, ∂F
where A=
1
∂u1 ∂F2 ∂u1
∂F1 ∂u2 ∂F2 ∂u2
Ut + AUx = S,
and
0 S= . S2
(a) Find the matrix A and its eigenvalues for the inviscid Burgers equation. Show that the conservation laws are not hyperbolic. (b) For compressible flows of velocity u, define the density u1 and the momentum u2 = uu1 , F1 = uu1 = u2 , F2 = uu2 + P = u22 /u1 + P , where P = P (u1 ) is the pressure, write the conservation equation in the matrix form Ut + AUx = S, and find the matrix A, its eigenvalues, eigenvectors, and the Riemann invariants. (c) Write the conservation matrix equation for the n × n matrix A.
13 Tables of Integral Transforms
In this chapter, we provide a set of short tables of integral transforms of the functions that are either cited in the text or are in most common use in mathematical, physical, and engineering applications. For exhaustive lists of integral transforms, the reader is referred to Erdélyi et al. (1954), Campbell and Foster (1948), Ditkin and Prudnikov (1965), Doetsch (1970), Marichev (1983), Debnath (1995), Debnath and Bhatta (2007), and Oberhettinger (1972).
13.1 Fourier Transforms
√1 2π
f (x)
F (k) =
1
f (ax + b)
1 |a|
2
eibx f (ax)
1 F ( k−b ) |a| a
3
f (n) (x)
(ik)n F (k)
4
xn f (x)
in F (n) (k)
5
f (−x)
F (k)
6
f (x)
F (−k)
7
f (x) ∗ g(x) ∞ = √12π −∞ f (x − ξ)g(ξ) dξ
F (k)G(k)
∞ −∞
exp(−ikx)f (x) dx
exp( ibk )F ( ka ) a
L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1_13,
676
13 Tables of Integral Transforms f (x) 2
∞
√1 2π
F (k) =
−∞
8
1 x If ft (x) = √4πκt exp(− 4κt ), ft (x) ∗ fs (x)
Fs+t (k)
9
f (x)g(x)
F (k) ∗ G(k)
10
f (x) cos ax
1 [F (k 2
11
f (x) sin ax
1 [F (k 2i
12
F (x)
f (−k)
13
xn eiax
√1 in δ (n) (k 2π
14
χ[−a,a] (x) = H(a − |x|)
15
(1 −
16
exp(−a|x|),
17
x exp(−a|x|),
18
exp(−ax2 ),
19
(x2 + a2 )−1 ,
|x| )H(1 a
−
|x| ) a
20 21
a>0 a>0
x(x2 + a2 )−1 c if a ≤ x ≤ b, 0 otherwise
22
|x| exp(−a|x|),
23
sin ax x
24
exp{−x(a − iω)}H(x)
25
(a2 − x2 )− 2 H(a − |x|)
(
sin[b(x2 +a2 ) 2 ]
− a)
sin2 ( ak ) 2
+ k 2 )−1
2 )(−2aik)(a2 π
√1 2a
+ k 2 )−2
2
exp(− k4a )
π exp(−a|k|) a
2
π 2
( −ik ) exp(−a|k|) 2a
√ic 1 (e−ibk 2π k
2 (a2 π
2
1
1 (x2 +a2 ) 2
(
2 )a(a2 π
π
1
26
a>0
− a) − F (k + a)]
√a ( ak )−2 2π 2
a>0
− a) + F (k + a)]
2 sin ak ( k ) π
a>0
exp(−ikx)f (x) dx
− e−iak )
− k 2 )(a2 + k 2 )−2
H(a − |k|)
i √1 2π (ω−k+ia)
π 2
π 2
J0 (ak) √ J0 (a b2 − k 2 )H(b − |k|)
13.2 Fourier Sine Transforms f (x) √ 27
cos(b
F (k) = a2 −x2 )
1 (a2 −x2 ) 2
28
e−ax H(x),
29
√1
30
δ (n) (x − a),
31
exp(iax)
|x|
π
H(a − |x|)
2
exp(−a|x|),
−∞
exp(−ikx)f (x) dx
√ J0 (a b2 + k 2 )
√1 (a 2π
a>0
∞
√1 2π
− ik)(a2 + k 2 )−1 1
1
√1 (ik)n 2π
n = 0, 1, 2, . . .
√
exp(−iak)
δ(k − a)
2π
32
xn sgn x
33
1
34
x
35
xn
36
x−1
−i
37
x−n
−i
38
cos(ax2 ),
a>0
2 1 √ [cos( k4a ) 2 a
+ sin( k4a )]
39
sin(ax2 ),
a>0
2 1 √ [cos( k4a ) 2 a
− sin( k4a )]
√ √ √
n! 2 π (ik)n+1
2πδ(k) 2πiδ (k) 2πin δ (n) (k) π 2
π 2
sgn k n−1
[ (−ik) sgn k] (n−1)! 2
2
13.2 Fourier Sine Transforms f (x)
Fs (k) =
1
exp(−ax),
2
x exp(−ax),
3
xα−1 ,
4
√1 x
a>0 a>0
0<α<1
1
(a2 + k 2 )− 2 [a + (a2 + k 2 ) 2 ] 2
a>0
2 k(a2 π
∞ 2 π
0
sin(kx)f (x) dx
+ k 2 )−1
2 (2ak)(a2 π
+ k 2 )−2
2 −α k Γ (α) sin( πα ) π 2
√1 , k
k>0
677
678
13 Tables of Integral Transforms f (x)
5
xα−1 e−ax ,
α > −1, a > 0
∞ Fs (k) = π2 0 sin(kx)f (x) dx 2 Γ (α)r −α sin(αθ), where π 1
r = (a2 + k 2 ) 2 , θ = tan−1 ( ka ) 2 tan−1 ( ka ), k > 0 π
6
x−1 e−ax ,
7
x exp(−a2 x2 )
8
erfc(ax)
9
x(a2 + x2 )−1
π
10
x(a2 + x2 )−2
π
11
H(a − x),
a>0
2
k 2−3/2 ( ak3 ) exp(− 4a 2)
2 1 [1 π k
2
2
a>0
exp(−ak),
12
x
J0 (ax)
2 1 (1 π k 2
ππ
13
14
x(a2 + x2 )−1 J0 (bx), a > 0, b > 0 √ J0 (a x),
2
a>0
sin−1 ( ka )
1
15
(x2 − a2 )ν− 2 H(x − a), |ν| < 12
16
x1−ν (x2 + a2 )−1 Jν (ax), ν − 32 , a, b > 0
17
x−ν Jν+1 (ax),
18
erfc(ax)
19
x−α ,
20
(ax − x2 )α− 2 H(a − x), α > − 12
ν > − 12
1
a
2
cos( a4k )
1
2ν− 2 ( ak )ν Γ (ν + 12 )J−ν (ak) π
a−ν exp(−ak)Iν (ab), a
k(a2 −k2 ) ν− 1 2 2
0<α<2
if 0 < k < a, if a < k < ∞
e−ak I0 (ab),
2 1 π k
(a > 0)
− cos ak)
2
π
a>0
k ( 2a ) exp(−ak),
−1
2
k − exp(− 4a 2 )]
ν− 1 2
aν+1 Γ (ν+ 1 ) 2
2 1 [1 π k
H(a − k) 2
k − exp(− 4a 2 )]
Γ (1 − α)k α−1 cos( απ ) 2 √
2Γ (α + 12 )( ak )α sin( ak )Jα ( ak ) 2 2
13.3 Fourier Cosine Transforms
13.3 Fourier Cosine Transforms
f (x)
Fc (k) =
1
exp(−ax),
a>0
(
2
x exp(−ax),
3
exp(−a2 x2 )
4
H(a − x)
5
xa−1 ,
6
cos(ax2 )
7
sin(ax2 ),
8
(a2 − x2 )ν− 2 H(a − x),
9
(a2 + x2 )−1 J0 (bx),
a>0
(
0
a>0 1
ν > − 12
2 )(a2 π
cos(kx)f (x) dx
+ k 2 )−1
− k 2 )(a2 + k 2 )−2 2
k exp(− 4a 2)
2 sin ak ( k ) π 2 Γ (a)k−a π
cos( aπ ) 2
2 1 √ [cos( k4a ) 2 a
+ sin( k4a )]
2
2 1 √ [cos( k4a ) 2 a
− sin( k4a )]
2
1
2ν− 2 Γ (ν + 12 )( ka )ν Jν (ak) a−1 exp(−ak)I0 (ab), b
(a2 −k2 )
10
x−ν Jν (ax),
11
(x2 + a2 )− 2 exp[−b(x2 + a2 ) 2 ]
12
xν−1 e−ax ,
ν > − 12
ν− 1 2 2 1
ν > 0, a > 0
0
π
a, b > 0
1
π
2 )a(a2 π
1 √ a 2
∞ 2
ν− 1 2
H(a−k)
aν Γ (ν+ 1 ) 2 1
K0 [a(k 2 + b2 ) 2 ], a > 0, b > 0 2 Γ (ν)r−ν cos nθ, where π 1
r = (a2 + k 2 ) 2 , θ = tan−1 ( ka ) 2 tan−1 ( k22 ) π
e−x sin x
13
2 x
14
sin[a(b2 − x2 ) 2 H(b − x)]
1
π 2
1
(ab)(a2 + k2 )− 2 1 2
× J1 [b(a2 + k 2 ) ] π
15
(1−x2 ) (1+x2 )2
16
x−α ,
17
( a1 + x)e−ax
2
0<α<1
k exp(−k)
π kα−1 2 Γ (α)
π
sec( πα ) 2
2a2 2 (a2 +k2 )2
679
680
13 Tables of Integral Transforms f (x)
Fc (k) =
log(1 +
19
+x log( ab2 +x 2 ),
20
a(x2 + a2 )−1 ,
2
2
π
0
a>0
√ (1−e−ak ) 2π k
a, b > 0
√ (e−bk −e−ak ) 2π k
a2 ), k2
18
∞ 2
a>0
π 2
exp(−ak),
cos(kx)f (x) dx
k>0
13.4 Laplace Transforms
∞
f (t)
f (s) =
1
e−at f (t)
f¯(s + a)
2
f (at)
1 ¯ s f(a) |a|
3
f (n) (t)
sn f (s) −
0
exp(−st)f (t) dt
n
sn−r f (r−1) (0)
r=1
4
t 0
1 ¯ f (s) s
f (τ ) dτ
n
5
tn f (t)
d (−1)n ds n f (s)
6
f (t − a)H(t − a)
exp(−as)f (s)
7
(f ∗ g)(t) =
8
tn−1 (n−1)!
9
t 0
∗f (t) =
f (t − τ )g(τ ) dτ t 0
(t−τ )n−1 f (τ ) dτ (n−1)!
[t] = greatest integer = largest integer that, is less than or equal to t
f (s)g(s) f¯(s) sn 1 (es s
(1 − e−as )−1
10
f (t) is periodic with period a
11
f (t) = H(t) − 2H(t − a) + 2H(t − 2a) − 2H(t − 3a) + · · ·
1 s
12
tn ,
n! sn+1
n = 0, 1, 2, 3, . . .
− 1)−1
tanh( as ) 2
a 0
e−st f (t) dt
13.4 Laplace Transforms ∞
f (t)
f (s) =
13
eat
1 s−a
14
tn e−at
Γ (n+1) (s+a)n+1
15
ta ,
16
eat cos bt
s−a (s−a)2 +b2
17
eat sin bt
b (s−a)2 +b2
18
1 √ t
a > −1
0
Γ (a+1) sa+1
π s
19
√ 2 t
20
t−1/2 exp(− at )
21
t−3/2 exp(− at )
22
√1 (1 πt
23
√1 (ebt 2 πt3
1 s
+ 2at)eat − eat )
π s
π s
π a
√ exp(−2 as) √ exp(−2 as)
s√ (s−a) s−a
√
s−a−
24
√ exp(a2 t) erf(a t)
√
a s(s−a2 )
25
√ exp(a2 t) erfc(a t)
√
1 √ s( s+a)
26
√1 πt
√ + a exp(a2 t) erf(a t)
√ s (s−a2 )
27
√1 πt
√ − a exp(a2 t) erfc(a t)
√1 s+a
28
exp(−at) √ b−a
29
exp(−st)f (t) dt
erf(
(b − a)t)
√ 1 iωt e [exp(−λz) erfc(ζ − iωt) 2 √ + exp(λz) erfc(ζ + iωt)] √ where ζ = z/2 νt, λ = iω ν
30
1 √ ) [exp(−ab) erfc( b−2at 2 2 t √ )] + exp(ab) erfc( b+2at 2 t
31
J0 (at)
√ s−b
1√ (s+a) s+b
(s − iω)−1 exp(−z
1
exp[−b(s + a2 ) 2 ]
1
(s2 + a2 )− 2
s ν
)
681
682
13 Tables of Integral Transforms ∞
f (t)
f (s) =
32
I0 (at)
(s2 − a2 )− 2
33
tα−1 exp(−at),
34
t−1 Jν (at)
35
√ J0 (a t)
1 s
36
√ ( a2 )ν tν/2 Jν (a t)
s−(ν+1) exp(− a4s ),
0
exp(−st)f (t) dt 1
Γ (α)(s + a)−α
α>0
√ ν −1 aν ( s2 + a2 + s)−ν , Re ν > − 12 2
exp(− a4s ) 2
√ exp(−a s),
2
37
a √ 2t πt
38
√1 πt
39
exp(− a 4t )
40
(t2 − a2 )− 2 H(t − a)
K0 (as),
41
δ (n) (t − a),
sn exp(−as)
42
tmα+β−1 Eα,β (± at), m = 0, 1, 2, . . .
43 44
exp(− a4t ) 2
exp(− a4t )
√1 s √
2 2
π a
1
n = 0, 1, . . .
√ exp(−a s), 2
√ π ( t )ν Jν (at) Γ (ν+ 1 ) 2a 2
1
(s+c)
√
√
exp(−a
s+b)
(s+b)
c)t}
√
√ exp(−a s+b) (s+c)
c}
− exp(a b − c) √ × erfc{ √a4t + t b − c}]
46
e−bt [
4t π
2
exp(− a4t )
− a erfc( √a4t )]
a≥0
a>0
(s2 + a2 )−(ν+ 2 ) ,
√
1 −ct e [exp(−a b − c) 2 √ × erfc{ √a4t − t b − √
a≥0
exp( as 2 ) erfc( as ),
c) − exp(a b − × erfc{ √a4t + (b − c)t}]
45
a>0
m!sα−β (sα ∓a)m+1
(m)
1 −ct e [exp(−a b − c) 2 × erfc{ √a4t − (b − √
Re ν > − 12
√ exp(−a s+b) (s+b)3/2
Re ν > − 12
13.5 Hankel Transforms f (t) 47
f (s) =
√ exp(ab + a2 t) erfc(a t +
49
Ln (t) =
50
n!√ H ( (2n)! πt 2n
51
0
exp(−st)f (t) dt
√ exp(−a s+b) (s+b)2
e−bt [(t + 12 a2 ) erfc( √a4t ) 2 2 − taπ exp(− a4t )]
48
∞
2
b √
) t
et dn (tn e−t ) n! dtn
√ −b s
1 (1 s
√ t)
b≥0
√e √ , s(a+ s)
− 1s )n
(1−s)n s
n+ 1 2
1 [f¯(s) 2πi
f (t)g(t)
∗ g¯(s)]
13.5 Hankel Transforms ∞
f (r)
Order n
f˜n (κ) =
1
f (ar)
0
1 ˜ κ f ( ), a2 n a
2
f (r)
1
−κf˜0 (κ)
3
∇2 f =
0
−κ2 f˜0 (κ)
4
(∇2 −
>−1
−κ2 f˜n (κ)
5
H(a − r)
0
a J (aκ) κ 1
6
exp(−ar)
0
a(a2 + κ2 )− 2
7
1 r
0
(a2 + κ2 )− 2
8
(a2 − r2 )H(a − r)
0
4a J (aκ) κ3 1
9
a(a2 + r2 )− 2
0
exp(−aκ)
1 r
0
(κ2 − a2 )− 2 H(κ − a)
10
1 d df ( ) r dr dr n2 )f r2
exp(−ar),
a>0
3
cos(ar)
0
rJn (κr)f (r) dr
a>0
3
1
−
1
2a2 J (aκ) κ2 0
683
684
13 Tables of Integral Transforms ∞
f (r)
Order n
f˜n (κ) =
11
1 r
0
(a2 − κ2 )− 2 H(a − κ)
12
1 (1 r2
0
cosh−1 ( κa )H(a − κ)
13
1 J (ar) r 1
0
1 H(a a
14
Y0 (ar)
0
( π2 )(a2 − κ2 )−1
15
K0 (ar)
0
(a2 + κ2 )−1
16
δ(r) r
0
1
17
δ(r − a)
>−1
aJn (aκ)
18
(r2 + b2 )− 2
0
(κ2 + a2 )− 2 exp{−b(κ2 + a2 ) 2 }
sin(ar) − cos ar)
1
0
rJn (κr)f (r) dr
1
− κ),
a>0
1
1
1
× exp{−a(r2 + b2 ) 2 } 1
19
(r2 + a2 )− 2
0
1 κ
20
exp(−ar)
1
κ(a2 + κ2 )−3/2
21
sin ar r
1
22
1 r
1
1 [1 κ
23
1 r2
1
1 [(κ2 κ
24
rn H(a − r)
>−1
1 n+1 a Jn+1 (aκ) κ
25
rn exp(−ar),
>−1
2n+1 Γ (n+ 3 )aκn 2 √1 π n+ 3 2 (a2 +κ2 )
26
rn exp(−ar 2 )
>−1
κn (2a)n+1
27
ra−1
>−1
28
rn (a2 − r2 )m−n−1 × H(a − r)
>−1
2m−n−1 Γ (m − n)am κn−m Jm (aκ)
29
rm exp(−r2 /a2 )
>−1
κn am+n+2 Γ (1 + m + n2 ) 2 2n+1 Γ (n+1) m n × 1 F1 (1+ 2 + 2 ; n+1;
exp(−ar) exp(−ar)
Re a > 0
exp(−aκ)
aH(κ−a) 1
κ(κ2 −a2 ) 2
−
a 1
(κ2 +a2 ) 2
]
1
+ a2 ) 2 − a]
2
exp(− κ4a )
2aΓ [ 1 (a+n+1)] 2
κa+1 Γ [ 1 (1−a+n)] 2
− 14 a2 κ2 )
13.5 Hankel Transforms
30 31
∞
f (r)
Order n
f˜n (κ) =
1 J (ar) r n+1
>−1
κn a−(n+1) H(a − κ),
>−1
2m an Γ (m + 1)( κa )m+1 × Jn+m+1 (aκ) 1 κ n ( ) if 0 < κ ≤ a, 2n a
rn (a2 − r2 )m H(a − r), m > −1
32
1 J (ar) r2 n
> 12
33
rn , (a2 +r 2 )m+1
34
rJn (κr)f (r) dr
0
a>0
if a < κ < ∞
1 a n ( ) 2n κ n−m
>−1
( κ2 )m Γa(m+1) Kn−m (aκ)
exp(−p2 r 2 )Jn (ar)
>−1
+κ aκ (2p2 )−1 exp(− a 4p 2 )In ( 2p2 )
35
1 r
>−1
36
rn (r 2 +a2 )n+1
a>0
2
2
1
37
exp(−ar)
rn (a2 −r 2 )
n+ 1 2
H(a − r)
((κ2 +a2 ) 2 −a)n 1
κn (κ2 +a2 ) 2
>−1
0 (aκ) ( κ2 )n ΓK(n+1)
<1
√1 ( κ )n Γ ( 1 2 π 2
1
1 [1 κ
− n)( sinκaκ ) 2
38
r−1 exp(−ar2 )
39
r−1 sin(ar 2 ),
a>0
1
1 κ
40
r−1 cos(ar2 ),
a>0
1
1 − cos( κ4a )
41
exp(−ar),
>−1
(a+n κ2 +a2 ) ( √κ2 2 )n (κ2 +a2 )3/2 a+ a +κ
42
exp(−ar 2 )J0 (br)
0
a 2
43
H(a−r) √
0
− exp(− κ4a )] 2
sin( κ4a ) 2
√
44
a2 −r 2
r n H(a−r)
√
a2 −r2
a>0
>−1
2
2
−b bκ exp(− κ 4a )I0 ( 2a )
aπ 2κ
J 1 (aκ), 2
a>0
π n+ 1 a 2 Jn+1 (aκ), 2κ
45
r−2 sin r
0
sin−1 ( κ1 ),
46
exp(−ar 2 )
0
1 2a
47
r exp(−ar 2 )
0
κ 4a2
κ>1 2
exp(− κ4a ) 2
exp(− κ4a )
a>0
685
686
13 Tables of Integral Transforms f (r)
48
If a = σ + iω in 46, f (r) = e
−σr2
f˜n (κ) =
0
2Ω exp(−σΩκ2 ) × {σ cos(ωΩκ2 ) + ω sin(ωΩκ2 )}, where Ω = [4(σ 2 + ω 2 )]−1
2
cos(ωr )
2
∞
Order n
0
rJn (κr)f (r) dr
49
f (r) = e−σr sin(ωr 2 )
0
2Ω exp(−σΩκ2 ) × {−ω cos(ωΩκ2 ) + σ sin(ωΩκ2 )}, where Ω = [4(σ2 + ω2 )]−1
50
δ(ar − b) = a>0
>−1
b J ( bκ ), a2 n a
51
δ (m) (ar − b),
> −1
(−1)m κm−1 a2 (m) × [ ab Jn ( bκ ) a
1 δ(r a
− ab ),
a>0
a>0
(m−1) bκ ( a )]
+ mJn
52
rn (r 2 +a2 )m+1
>−1
( κ2 )m Γa(m+1) Kn−m (aκ)
n−m
53
rm−n Jm (ar)
>−1
am 2m−n+1 (κ2 −a2 )n−m−1 , κn Γ (n−m)
κ>a
13.6 Finite Hankel Transforms
f (r) where c is a constant
a
Order n
f˜n (ki ) =
0
( ac )J1 (aki ) ki
0
4a J (aki ) ki3 1
0
rJn (rki )f (r) dr
1
c,
2
(a2 − r2 )
3
(a2 − r2 )− 2
0
ki−1 sin(aki )
4
J0 (αr) J0 (αa)
0
i − (α2ak J (aki ) −k2 ) 1
5
1 r
1
ki−1 {1
6
r−1 (a2 − r2 )− 2
1
(1−cos aki ) (aki )
7
rn
>−1
an+1 Jn+1 (aki ) ki
8
Jν (αr) Jν (αa)
>−1
aki J (aki ) (α2 −ki2 ) ν
1
i
1
− J0 (aki )}
13.6 Finite Hankel Transforms
9
a
Order n
f˜n (ki ) =
>−1
π {J n2 ( ak2 i )}2 2
1
< 12
Γ(1 −n) √2 n k n−1 i π2
1
>− 12
f (r) 1
r−n (a2 − r2 )− 2
10
rn (a2 − r2 )−(n+ 2 )
11
rn−1 (a2 − r2 )n− 2
√
π Γ (n 2
0
rJn (rki )f (r) dr
sin(aki )
+ 12 )( k2i )n a2n Jn2 ( ak2 i )
687
A Some Special Functions and Their Properties
The main purpose of this appendix is to introduce several special functions and to state their basic properties that are most frequently used in the theory and applications of ordinary and partial differential equations. The subject is, of course, too vast to be treated adequately in so short a space, so that only the more important results will be stated. For a fuller discussion of these topics and of further properties of these functions the reader is referred to the standard treatises on the subject.
A-1 Gamma, Beta, and Error Functions The gamma function (also called the factorial function) is defined by a definite integral in which a variable appears as a parameter ∞ Γ (x) = e−t tx−1 dt, x > 0. (A-1.1) 0
The integral (A-1.1) is uniformly convergent for all x in [a, b] where 0 < a ≤ b < ∞, and hence, Γ (x) is a continuous function for all x > 0. Integrating (A-1.1) by parts, we obtain the fundamental property of Γ (x) ∞ ∞ e−t tx−2 dt Γ (x) = −e−t tx−1 0 + (x − 1) 0
= (x − 1)Γ (x − 1)
for x − 1 > 0.
Then we replace x by x + 1 to obtain the fundamental result Γ (x + 1) = x Γ (x).
(A-1.2)
In particular, when x = n is a positive integer, we make repeated use of (A-1.2) to obtain Γ (n + 1) = nΓ (n) = n(n − 1)Γ (n − 1) = · · · = n(n − 1)(n − 2) · · · 3 · 2 · 1Γ (1) = n!, where Γ (1) = 1. L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1,
(A-1.3)
690
A Some Special Functions and Their Properties
We put t = u2 in (A-1.1) to obtain ∞ exp −u2 u2x−1 du, Γ (x) = 2
x > 0.
(A-1.4)
0
Letting x = 12 , we find √ ∞ 2 1 π √ exp −u du = 2 Γ =2 = π. 2 2 0
(A-1.5)
Using (A-1.2), we deduce √ π 1 3 1 = Γ = Γ . 2 2 2 2
(A-1.6)
). Similarly, we can obtain the values of Γ ( 52 ), Γ ( 72 ), . . . , Γ ( 2n+1 2 The gamma function can also be defined for negative values of x by the rewritten form of (A-1.2) as Γ (x) =
Γ (x + 1) , x
x = 0, −1, −2, . . . .
(A-1.7)
For example, √ Γ ( 12 ) 1 1 = −2 Γ Γ − = = −2 π, 1 2 2 −2 1 Γ (− 2 ) 3 4√ = Γ − = π. 3 2 3 −2 We differentiate (A-1.1) with respect to x to obtain ∞ d d x e−t Γ (x) = Γ (x) = t dt dx dx t 0 ∞ ∞ e−t d = tx−1 (log t)e−t dt. exp(x log t) dt = dx t 0 0 At x = 1, this gives
∞
Γ (1) =
e−t log t dt = −γ,
(A-1.8) (A-1.9)
(A-1.10)
(A-1.11)
0
where γ is called the Euler constant and has the value 0.5772. The graph of the gamma function is shown in Figure A.1. The volume, Vn , and the surface area, Sn , of a sphere of radius r in n-dimensional space Rn are given by Vn = Thus,
dVn dr
= Sn .
{Γ ( 12 )}n r n , Γ ( n2 + 1)
Sn =
2{Γ ( 12 )}n r n−1 . Γ ( n2 )
A-1 Gamma, Beta, and Error Functions
691
Fig. A.1 The gamma function.
In particular, when n = 2, 3, . . . , we get V2 = πr2 , S2 = 2πr; V3 = S3 = 4πr2 ; etc. Using (A-1.2) and (A-1.5), we obtain the following results:
4 πr3 , 3
π m r2m 2π m r 2m−1 , S2m = , m! (m − 1)! 2(2π)m r2m+1 22m+1 m!π m r 2m , S2m+1 = . = 1.3.5 · · · (2m + 1) (2m)!
V2m = V2m+1
Legendre Duplication Formula Several useful properties of the gamma function are recorded below for reference without proof. We begin with √ 1 22x−1 Γ (x)Γ x + = πΓ (2x). (A-1.12) 2 In particular, when x = n (n = 0, 1, 2, . . .), √ π (2n)! 1 = . Γ n+ 2 22n n!
(A-1.13)
The following properties also hold for Γ (x): Γ (x)Γ (1 − x) = π cosec ∞ πx, x is a noninteger, Γ (x) = px exp(−pt) tx−1 dt, ∞0 exp xt − et dt. Γ (x) = −∞
(A-1.14) (A-1.15) (A-1.16)
692
A Some Special Functions and Their Properties
Γ (x + 1) ∼ n! ∼
√ √
1
2π exp(−x)xx+ 2
for large x,
(A-1.17)
n+ 12
for large n.
(A-1.18)
2π exp(−n)x
The latter formulas are known as Stirling approximation of Γ (x + 1) for large x and of n! for large n. The incomplete gamma function, γ(x, a), is defined by the integral x e−t ta−1 dt, a > 0. (A-1.19) γ(a, x) = 0
The complementary incomplete gamma function, Γ (a, x), is defined by the integral ∞ Γ (a, x) = e−t ta−1 dt, a > 0. (A-1.20) x
Thus, it follows that γ(a, x) + Γ (a, x) = Γ (a). The beta function, denoted by B(x, y), is defined by the integral t tx−1 (1 − t)y−1 dt, x > 0, y > 0. B(x, y) =
(A-1.21)
(A-1.22)
0
The beta function B(x, y) is symmetric with respect to its arguments x and y, that is, B(x, y) = B(y, x). (A-1.23) This follows from (A-1.22) by the change of variable 1 − t = u, that is, 1 B(x, y) = uy−1 (1 − u)x−1 du = B(y, x). 0
If we make the change of variable t = u/(1 + u) in (A-1.22), we obtain another integral representation of the beta function ∞ ∞ x−1 −(x+y) B(x, y) = u (1 + u) du = uy−1 (1 + u)−(x+y) du. (A-1.24) 0
0
Putting t = cos2 θ in (A-1.22), we derive
π/2
cos2x−1 θ sin2y−1 θ dθ.
B(x, y) = 2
(A-1.25)
0
Several important results are recorded below for ready reference without proof: 1 1 = π, (A-1.26) , B(1, 1) = 1, B 2 2 x−1 B(x, y) = B(x − 1, y), (A-1.27) x+y−1
A-1 Gamma, Beta, and Error Functions
693
Fig. A.2 The error function and the complementary error function.
Γ (x)Γ (y) , B(x, y) = Γ (x + y) 1+x 1−x πx B = π sec , , 2 2 2
(A-1.28) 0 < x < 1.
The error function, erf(x), is defined by the integral x 2 exp −t2 dt, −∞ < x < ∞. erf(x) = √ π 0
(A-1.29)
(A-1.30)
Clearly, it follows from (A-1.30) that erf(−x) = − erf(x), d 2 erf(x) = √ exp −x2 , dx π erf(0) = 0, erf(∞) = 1. The complementary error function, erfc(x), is defined by the integral ∞ 2 exp −t2 dt. erfc(x) = √ π x
(A-1.31) (A-1.32) (A-1.33)
(A-1.34)
Clearly, it follows that erfc(x) = 1 − erf(x), erfc(0) = 1, erfc(∞) = 0.
(A-1.35) (A-1.36)
The graphs of erf(x) and erfc(x) are shown in Figure A.2. 1 erfc(x) ∼ √ exp −x2 x π
for large x.
(A-1.37)
Closely associated with the error function are the Fresnel integrals, which are defined by
694
A Some Special Functions and Their Properties
Fig. A.3 The Fresnel integrals C(x) and S(x).
x
C(x) = 0
πt2 cos 2
dt
and
x
S(x) = 0
πt2 sin 2
dt.
(A-1.38)
These integrals arise in diffraction problems in optics, in water waves, in elasticity, and elsewhere. Clearly, it follows from (A-1.38) that C(0) = 0 = S(0), π C(∞) = S(∞) = , 2 2 πx d , C(x) = cos dx 2
(A-1.39) d πx2 . S(x) = sin dx 2
(A-1.40) (A-1.41)
It also follows from (A-1.38) that C(x) has extrema at the points where x2 = (2n + 1), n = 0, 1, 2, 3, . . . , and S(x) has extrema at the points where x2 = 2n, n = 1, 2, √3, . . . . The largest maxima occur first and are found to be C(1) = 0.7799 and S( 2) = 0.7139. We also infer that both C(x) and S(x) are oscillatory about the line y = 0.5. The graphs of C(x) and S(x) for non-negative real x are shown in Figure A.3. We prove further properties of the Gamma and the Beta functions. We first prove that π/2 1 sin2p−1 x cos2q−1 x dx = B(p, q). (A-1.42) 2 0 We put sin2 x = t so that the left hand side of the above integral becomes 1 π/2 2p−2 sin x cos2q−2 x · 2 cos x sin x dx 2 0 1 1 1 p−1 t (1 − q)q−1 dt = B(p, q). (A-1.43) = 2 0 2 We next prove that
A-1 Gamma, Beta, and Error Functions
1 1 = 22p−1 Γ (p) Γ p + . 2 2
Γ (2p)Γ We have π/2 sin2p 2x dx = 0
Putting q =
1 2
1 2
π
π/2
sin2p θ dθ = 2 0
0
1 sin2p θ dθ, 2
695
(A-1.44)
(2x = θ).
in (A-1.42) with x = 2θ gives
π/2 π/2 1 1 2p =2 sin x dx = 2 sin2p 2θ dθ B p+ , 2 2 0 0 π/2 sin2p θ cos2p θ dθ = 22p+1 0 1 1 = 22p B p + , p + , 2 2 which is, using (A-1.28) and (A-1.2), Γ (p + 12 )Γ ( 12 ) Γ (p + 12 )Γ (p + 12 ) Γ (p + 12 )Γ (p + 12 ) = 22p = 22p , Γ (p + 1) Γ (2p + 1) 2p Γ (2p) or equivalently, 1 1 2p−1 Γ (2p) Γ =2 . Γ (p)Γ p + 2 2 We next define
f (n, t) =
(1 − nt )n tx−1
if 0 ≤ t ≤ n,
0
if t ≥ n.
(A-1.45)
Using n x = e−x , lim 1 − n→∞ n we obtain, for fixed t, lim f (n, t) = lim
n→∞
n→∞
t 1− n
n
tx−1 = e−t tx−1 .
Hence, for x > 0, Γ (x) =
∞
−t x−1
∞
e t dt = lim f (n, t) dt n→∞ 0 n0 n t 1− tx−1 dt, = lim n→∞ 0 n
(A-1.46)
696
A Some Special Functions and Their Properties
which is, putting
t n
= z,
1
(1 − z)n z x−1 dz
= lim nx n→∞
= = = =
0
1
1 zx n−1 z dz +n (1 − z) (1 − z) · lim n n→∞ x 0 x 0 1 n lim nx · (1 − z)n−1 z x dz n→∞ x 0 1 n · (n − 1) · · · 1 z n+x−1 dz lim nx n→∞ x · (x + 1) · · · (n + x − 1) 0 nx n! . lim n→∞ x(x + 1) · · · (x + n)
x
n
(A-1.47)
This is the celebrated Gauss formula. We next prove that, for 0 < x < 1, Γ (x)Γ (1 − x) =
π . sin πx
(A-1.48)
Since x and x − 1 are positive and not integers, we use the Gauss formula (A-1.47) so that nx n! n1−x n! n→∞ x(x + 1) · · · (x + n) (1 − x)(1 − x + 1) · · · (1 − x + n) (n!)2 n 1 · = lim n→∞ x (1 + x)(1 − x) · · · (n + x)(n − x)(n + 1 − x) 1 1 (n!)2 = lim · 2 n→∞ x (1 − x )(4 − x2 ) · · · (n2 − x2 ) {1 + 1−x } n 1 1 lim = x n→∞ (1 − x2 )(1 − x222 ) · · · (1 − nx22 ) ∞ −1 x2 1 1− 2 lim = , x n→∞ n=1 n
Γ (x)Γ (1 − x) = lim
which is, by the product formula for the sine function, =
xπ π 1 · = . x sin πx sin πx
Finally, we show that the (2n)th order moment of the standard normal probability density function 2 x 1 √ exp − (A-1.49) 2 2π is
A-2 Bessel and Airy Functions
1 2n . E X 2n = √ Γ n + π 2
697
(A-1.50)
We have
2 ∞ x 1 2n x exp − dx = √ E X 2 2π ∞ ∞ 2 2 x 2 x x2n exp − = dx, =t π 0 2 2 ∞ 2 1 (2t)n exp(−t)(2t)− 2 dt = π 0 1 n 1 1 n ∞ n− 1 −t 2 t e dt = √ 2 Γ n + = √ 2 . π π 2 0 √ Using Γ (x + 1) = xΓ (x) and Γ ( 12 ) = π, we obtain E X 2n = (2n − 1)(2n − 3) · · · 5 · 3 · 1.
2n
(A-1.51)
A random variable X with values in (0, 1) has the Beta distribution if its density function is, for some p, q > 0, 1 xp−1 (1 − x)q−1 , 0 < x < 1; (A-1.52) B(p, q) 1 n 1 B(n + p, q) . (A-1.53) E X = xn xp−1 (1 − x)q−1 dx = B(p, q) 0 B(p, q) f (x) =
When n = 1, E(X) =
B(p + 1, q) p = . B(p, q) p+q
(A-1.54)
A random variable with values in (0, ∞) has the Gamma distribution if, for some p > 0 and q > 0, f (x) =
q p p−1 −qx x e . Γ (p)
(A-1.55)
A-2 Bessel and Airy Functions The Bessel function of the first kind of order v (non-negative real number) is denoted by Jv (x) and defined by Jv (x) = xv This series is convergent for all x.
∞
(−1)r x2r . 22r+v r! Γ (r + v + 1) r=0
(A-2.1)
698
A Some Special Functions and Their Properties
The Bessel function y = Jv (x) satisfies the Bessel equation x2 y + xy + x2 − v 2 y = 0.
(A-2.2)
When v is not a positive integer or zero, Jv (x) and J−v (x) are two linearly independent solutions so that y = AJv (x) + BJ−v (x)
(A-2.3)
is the general solution of (A-2.2), where A and B are arbitrary constants. However, when v = n, where n is a positive integer or zero, Jn (x) and J−n (x) are no longer independent, but are related by the equation J−n (x) = (−1)n Jn (x).
(A-2.4)
Thus, when n is a positive integer or zero, equation (A-2.2) has only one solution given by n+2r ∞ (−1)r x . (A-2.5) Jn (x) = r!(n + r)! 2 r=0 A second solution, known as Neumann’s or Webber’s solution, Yn (x) is given by (A-2.6)
Yn (x) = lim Yv (x), v→n
where
(cos vπ)Jv (x) − J−v (x) . sin vπ Thus, the general solution of (A-2.2) is Yv (x) =
y(x) = A Jn (x) + B Yn (x),
(A-2.7)
(A-2.8)
where A and B are arbitrary constants. In particular, from (A-2.5), J0 (x) =
2r ∞ (−1)r x r=0 ∞
(r!)2
2
,
2r+1 x (−1)r J1 (x) = . r!(r + 1)! 2 r=0
(A-2.9) (A-2.10)
Clearly, it follows from (A-2.9) and (A-2.10) that J0 (x) = −J1 (x).
(A-2.11)
Bessel’s equation may not always arise in the standard form given in (A-2.2), but more frequently as (A-2.12) x2 y + xy + k2 x2 − v 2 y = 0
A-2 Bessel and Airy Functions
699
with the general solution (A-2.13)
y(x) = AJv (kx) + BYv (kx).
The recurrence relations are recorded below for easy reference without proof: v Jv (x) − Jv (x), (A-2.14) Jv+1 (x) = x v Jv (x) + Jv (x), Jv−1 (x) = (A-2.15) x 2v Jv (x), Jv−1 (x) + Jv+1 (x) = (A-2.16) x (A-2.17) Jv−1 (x) − Jv+1 (x) = 2Jv (x). We have, from (A-2.5), xn Jn (x) =
∞ (−1)r 2−(n+2r) r=0
r!(n + r)!
x2n+2r .
Differentiating both sides of this result with respect to x and using the fact that 2(n + r)/(n + r)! = 2/(n + r − 1)!, it turns out that ∞
(−1)r 2−(n+2r+1) 2n+2r−1 d n = xn Jn−1 (x). x Jn (x) = x dx r!(n + r − 1)! r=0
(A-2.18)
Similarly, we can show d −n x Jn (x) = −x−n Jn+1 (x). dx
(A-2.19)
The generating function for the Bessel function is ∞ 1 1 = tn Jn (x). exp x t − 2 t n=−∞
(A-2.20)
The integral representation of Jn (x) is 1 π Jn (x) = cos(nθ − x sin θ) dθ. π 0
(A-2.21)
The following are known as the Lommel integrals: a x Jn (px)Jn (qx) dx 0
=
(q 2
a p Jn (qa)Jn (pa) − q Jn (pa) Jn (qa) , 2 −p )
p = q,
(A-2.22)
700
A Some Special Functions and Their Properties
Fig. A.4 Graphs of y = J0 (x), J1 (x), and J2 (x).
and
a
x Jn2 (px) dx 0
When n = ± 12 , J 12 (x) =
a2 2 n2 2 J (pa) + 1 − 2 2 Jn (pa) . = 2 n p a
2 sin x, πx
J− 12 (x) =
2 cos x. πx
(A-2.23)
(A-2.24)
A rough idea of the shape of the Bessel functions when x is large may be obtained 1 from equation (A-2.2). Substitution of y = x− 2 u(x) eliminates the first derivative, and hence, gives the equation 4n2 − 1 u = 0. (A-2.25) u + 1 − 4x2 For large x, this equation approximately becomes u + u = 0.
(A-2.26)
This equation admits the solution u(x) = A cos(x + ε), that is, A y = √ cos(x + ε). x
(A-2.27)
This suggests that Jn (x) is oscillatory and has an infinite number of zeros. It also tends to zero as x → ∞. The graphs of Jn (x) for n = 0, 1, 2 and for n = ± 12 are shown in Figures A.4 and A.5, respectively. An important special case arises in particular physical problems when k 2 = −1 in equation (A-2.12). we then have the modified Bessel equation x2 y + xy − x2 + v 2 y = 0, (A-2.28) with the general solution y = AJv (ix) + BYv (ix).
(A-2.29)
A-2 Bessel and Airy Functions
701
Fig. A.5 Graphs of J 1 (x) and J− 1 (x). 2
2
Fig. A.6 Graphs of y = Y0 (x), Y1 (x), and Y2 (x).
We now define a new function Iv (x) = i−v Jv (ix),
(A-2.30)
and then use the series (A-2.1) for Jv (x) so that v+2r v+2r ∞ ∞ (−1)r 1 ix x Iv (x) = i−v = . r!Γ (r + v + 1) 2 r!Γ (r + v + 1) 2 r=0 r=0 (A-2.31) Similarly, we can find the second solution, Kv (x), of the modified Bessel equation (A-2.28). Usually, Iv (x) and Kv (x) are called modified Bessel functions and their properties can be obtained in a similar way to those of Jv (x) and Yv (x). The graphs of Y0 (x), Y1 (x), and Y2 (x) are shown in Figure A.6. We state a few important infinite integrals involving Bessel functions which arise frequently in the application of Hankel transforms. ∞ (2b)v Γ (v + 12 ) 1 exp(−at)Jv (bt)tv dt = √ , v>− , (A-2.32) v+ 12 2 2 2 π(a + b ) 0 ∞ 2a(2b)v Γ (v + 32 ) exp(−at)Jv (bt)tv+1 dt = √ v > −1, (A-2.33) 3 , π(a2 + b2 )v+ 2 0
702
A Some Special Functions and Their Properties
Fig. A.7 The Airy function.
∞ 0
∞ 0
exp −a2 t2 Jv (bt)tv+1 dt =
bv b2 exp − 2 , (2a2 )v+1 4a
v > −1,
(A-2.34) 2 2 2 1 b + c2 bc exp −a t Jv (bt)Jv (ct)t dt = 2 exp − Iv , 2a 4a2 2a2 v > −1, (A-2.35) ∞ 2μ−v−1 Γ (μ) 1 1 2 , 0<μ< , v>− . t2μ−v−1 Jv (t) dt = Γ (v − μ + 1) 2 2 0 (A-2.36)
The Airy function, y = Ai (x), is the first solution of the differential equation y − xy = 0.
(A-2.37)
The second solution is denoted by Bi (x). Then these functions are expressed in terms of the Bessel and modified Bessel functions in the form x 1 x 2 3/2 2 3/2 I− 13 x x K 1 (ξ), (A-2.38) − I 13 = Ai (x) = 3 3 3 π 3 3 x x iπ 2 3/2 2 3/2 1 1 Bi (x) = + I3 = I− 3 x x Re e 6 H 13 (−iξ) , 3 3 3 3 (A-2.39) where ξ = 23 x3/2 . The integral representation of Ai (x) is 1 3 1 ∞ cos t + x t dt. Ai (x) = π 0 3
(A-2.40)
The graph of y = Ai (x) is shown in Figure A.7 using the values of Ai (x) at x = 0 and x → ∞: 1 1 Ai (0) = √ Bi (0) = 3/2 2 = 0.355028, 3 Γ(3) 3 Ai (x), Bi (x) → [0, ∞] as x → ∞.
A-3 Legendre and Associated Legendre Functions
Similarly, the graph of y = Bi (x) can be drawn. The integral representation of Bi (x) is 1 ∞ t3 t3 Bi (x) = + sin xt + dt. exp xt − π 0 3 3 A slightly more general integral representation of Ai (ax) is ∞ t3 1 cos xt + 3 dt. Ai (ax) = πa 0 3a
703
(A-2.41)
(A-2.42)
When a = 1, this reduces to (A-2.40). The general power series solution of the Airy equation (A-2.37) is given by x3 x6 x3n + + ··· + + ··· y(x) = a0 1 + 2·3 2·3·5·6 2 · 3 · · · (3n − 1)(3n) x4 x7 x3n+1 + + ··· + + ··· + a1 x + 3·4 3·4·6·7 3 · 4 · · · (3n)(3n + 1) (A-2.43) ∞ 3n x = a0 1 + 3 · 4 · · · (3n − 4)(3n − 3)(3n − 1)(3n) n=1 ∞ x3n+1 , (A-2.44) + a1 x + 3 · 4 · · · (3n − 3)(3n − 2)(3n)(3n + 1) n=1 where a0 and a1 are arbitrary constants of integration. Finally, the asymptotic representations of Ai (x) and Bi (x) are given by 1 2 3/2 Ai (x) ≈ √ 1/4 exp − x as x → +∞, (A-2.45) 3 2 πx 3 1 as x → ∞. (A-2.46) Bi (x) ≈ √ 1/4 exp x3/2 2 πx
A-3 Legendre and Associated Legendre Functions The Legendre polynomials, Pn (x), are defined by the Rodrigues formula Pn (x) =
1 2n n!
The first seven Legendre polynomials are P0 (x) = 1, P1 (x) = x,
n dn 2 x −1 . n dx
(A-3.1)
704
A Some Special Functions and Their Properties
P2 (x) = P3 (x) = P4 (x) = P5 (x) = P6 (x) =
1 2 3x − 1 , 2 1 3 5x − 3x , 2 1 35x4 − 30x2 + 3 , 8 1 63x5 − 70x3 + 15x , 8 1 231x6 − 315x4 + 105x2 − 5 . 16
The generating function for the Legendre polynomials is ∞ − 1 tn Pn (x). 1 − 2xt + t2 2 =
(A-3.2)
n=0
This function provides more information about the Legendre polynomials. For example, Pn (−1) = (−1)n , 1 · 3 · 5 · · · (2n − 1) (2n − 1)!! = (−1)n , P2n (0) = (−1)n 2n n! (2n)!! P2n+1 (0) = 0, n = 0, 1, 2, . . . , dn (2n)! Pn (x) = n , Pn (−x) = (−1)n Pn (x), n dx 2 n!
(A-3.3)
Pn (1) = 1,
(A-3.4) (A-3.5) (A-3.6)
where the double factorial is defined by (2n − 1)!! = 1 · 3 · 5 · · · (2n − 1)
and
(2n)!! = 2 · 4 · 6 · · · (2n).
The graphs of the first four Legendre polynomials are shown in Figure A.8. The recurrence relations for the Legendre polynomials are (n + 1)Pn+1 (x) (x) − Pn−1 (x) Pn+1 2 1 − x Pn (x) 1 − x2 Pn (x)
= (2n + 1)xPn (x) − nPn−1 (x), = (2n + 1)Pn (x),
(A-3.7) (A-3.8)
= nPn−1 (x) − nxPn (x),
(A-3.9)
= (n + 1)x Pn (x) − (n + 1) Pn+1 (x).
(A-3.10)
The Legendre polynomials, y = Pn (x), satisfy the Legendre differential equation 1 − x2 y − 2x y + n(n + 1)y = 0. (A-3.11) If n is not an integer, both solutions of (A-3.11) diverge at x = ±1. The orthogonal relation is 1 2 δnm . Pn (x)Pm (x) dx = (2n + 1) −1
(A-3.12)
A-3 Legendre and Associated Legendre Functions
705
Fig. A.8 Graphs of y = P0 (x), P1 (x), P2 (x), and P3 (x).
The associated Legendre functions are defined by m dm m dm+n 2 n 1 1 − x2 2 x − 1 , (A-3.13) Pn (x) = n Pnm (x) = 1 − x2 2 m m+n dx 2 n! dx where 0 ≤ m ≤ n. Clearly, it follows that Pn0 (x) = Pn (x),
(A-3.14)
Pnm (−x) = (−1)n+m Pnm (x),
Pn−m (x) = (−1)m
(n − m)! m P (x). (A-3.15) (n + m)! n
The generating function for Pnm (x) is m
(2m)!(1 − x2 ) 2
m+ 12
2m m!(1 − 2tx + t2 )
=
∞
m Pr+m (x)tr .
(A-3.16)
r=0
The recurrence relations are m m (2n + 1)xPnm (x) = (n + m)Pn−1 (x) + (n − m + 1)Pn+1 (x), 1 d 2 1 − x2 2 P m (x) = Pnm+1 (x) − (n + m)(n − m + 1)Pnm−1 (x). dx n
(A-3.17) (A-3.18)
The associated Legendre functions Pnm (x) are solutions of the differential equation m2 y = 0. (A-3.19) 1 − x2 y − 2xy + n(n + 1) − (1 − x2 ) This reduces to the Legendre equation when m = 0.
706
A Some Special Functions and Their Properties
Listed below are a few associated Legendre functions with x = cos θ: 1 P11 (x) = 1 − x2 2 = sin θ, 1 P21 (x) = 3x 1 − x2 2 = 3 cos θ sin θ, P22 (x) = 3 1 − x2 = 3 sin2 θ, 1 3 3 P31 (x) = 5x2 − 1 1 − x2 2 = 5 cos2 θ − 1 sin θ, 2 2 P32 (x) = 15x 1 − x2 = 15 cos θ sin2 θ, 3/2 = 15 sin3 θ. P33 (x) = 15 1 − x2 The orthogonal relations are
1 −1
1 −1
1 − x2
Pnm (x)Ptm (x) dx =
−1
Pnm (x)Pn (x) dx =
( + m)! 2 · δn , (2 + 1) ( − m)!
(A-3.20)
(n + m)! δm . m(n − m)!
(A-3.21)
A-4 Jacobi and Gegenbauer Polynomials (α,β)
The Jacobi polynomials, Pn mula Pn(α,β) (x) =
(x), of degree n are defined by the Rodrigues for-
(−1)n dn (1 − x)−α (1 + x)−β n (1 − x)α+n (1 + x)β+n , (A-4.1) n 2 n! dx
where α > −1 and β > −1. When α = β = 0, the Jacobi polynomials become Legendre polynomials, that is, (A-4.2) Pn (x) = Pn(0,0) (x), n = 0, 1, 2, . . . . On the other hand, the associated Laguerre functions arise as the limit 2x α (α,β) Ln (x) = lim Pn . 1− β→∞ β (α,β)
The recurrence relations for Pn
(A-4.3)
(x) are (α,β)
2(n + 1)(α + β + n + 1)(α + β + 2n)Pn+1 (x) = (α + β + 2n + 1) α2 − β 2 + x(α + β + 2n + 2)(α + β + 2n) Pn(α,β) (x) (α,β)
− 2(α + n)(β + n)(α + β + 2n + 2)Pn−1 (x), where n = 1, 2, 3, . . . , and
(A-4.4)
A-4 Jacobi and Gegenbauer Polynomials (α,β)
Pn(α,β−1) (x) − Pn(α−1,β) (x) = Pn−1 (x).
707
(A-4.5)
The generating function for Jacobi polynomials is 2(α+β) R−1 (1 − t + R)−α (1 + t + R)−β =
∞
Pn(α,β) (x)tn ,
(A-4.6)
n=0 1
where R = (1 − 2xt + t2 ) 2 . (α,β) The Jacobi polynomials, y = Pn (x), satisfy the differential equation 1 − x2 y + (β − α) − (α + β + 2)x y + n(n + α + β + 1)y = 0. (A-4.7) The orthogonal relation is
1
(1 − x) (1 + x) α
−1
where δn =
β
(α,β) Pn(α,β) (x)Pm (x) dx
=
0 δn
if n = m, if n = m,
2α+β+1 Γ (n + α + 1)Γ (n + β + 1) . n!(α + β + 2n + 1)Γ (α + β + n + 1)
(A-4.8)
(A-4.9)
When α = β = v − 12 , the Jacobi polynomials reduce to the Gegenbauer polynomials, Cnv (x), which are defined by the Rodrigues formula Cnv (x) =
1 1 n (−1)n 2 v− 2 d 2 v+n− 2 1 − x 1 − x . 2n n! dxn
(A-4.10)
The generating function for Cnv (x) of degree n is
1 − 2x t + t2
−v
=
∞ n=0
Cnv (x) tn ,
1 |t| < 1, |x| ≤ 1, v > − . 2
(A-4.11)
The recurrence relations are v v (x) − 2(v + n)xCnv (x) + (2v + n − 1)Cn−1 (x) = 0, (A-4.12) (n + 1)Cn+1 v+1 v (x) − 2vCnv+1 (x) + 2vCn−1 (x) = 0, (n + 1)Cn+1 d v+1 C v (x) = 2vCn+1 (x). dx n
The differential equation satisfied by y = Cnv (x) is 1 − x2 y − (2v + 1)xy + n(n + 2v)y = 0.
(A-4.13) (A-4.14)
(A-4.15)
The orthogonal property is
1 −1
v− 12 v v Cn (x)Cm (x) dx = δn δnm , 1 − x2
(A-4.16)
708
A Some Special Functions and Their Properties
where δn =
21−2v nΓ (n + 2 v) . n!(n + v)[Γ (v)]2
(A-4.17)
When v = 12 , the Gegenbauer polynomials reduce to Legendre polynomials, that is, 1
(A-4.18)
Cn2 (x) = Pn (x).
The Hermite polynomials can also be obtained from the Gegenbauer polynomials as the limit x (A-4.19) Hn (x) = n! lim v −n/2 Cnv √ . v→∞ v Finally, when α = β = 12 , the Gegenbauer polynomials reduce to the well-known Chebyshev polynomials, Tn (x), which are defined by a solution of the second order difference equation un+2 − 2x un+1 + un = 0, |x| ≤ 1, u(0) = u0 and u(1) = u1 .
(A-4.20) (A-4.21)
The generating function for Tn (x) is ∞ (1 − t2 ) = T (x) + 2 Tn (x)tn , 0 (1 − 2x t + t2 ) n=1
|x| ≤ 1, t < 1.
(A-4.22)
The first seven Chebyshev polynomials of degree n of the first kind are T0 (x) = 1, T1 (x) = x, T2 (x) = 2x2 − 1, T3 (x) = 4x3 − 3x, T4 (x) = 8x4 − 8x2 + 1, T5 (x) = 16x5 − 20x3 + 5x, T6 (x) = 32x6 − 48x4 + 18x2 − 1. The graphs of the first four Chebyshev polynomials are shown in Figure A.9. The Chebyshev polynomials y = Tn (x) satisfy the differential equation (A-4.23) 1 − x2 y − xy + n2 y = 0. It follows from (A-4.22) that Tn (x) satisfies the recurrence relations Tn+1 (x) − 2xTn (x) + Tn−1 (x) = 0,
(A-4.24)
Tn+m (x) − 2Tn (x)Tm (x) + Tn−m (x) = 0, 1 − x2 Tn (x) + nxTn (x) − nTn−1 (x) = 0.
(A-4.25) (A-4.26)
A-4 Jacobi and Gegenbauer Polynomials
709
Fig. A.9 Chebyshev polynomials y = Tn (x).
The parity relation for Tn (x) is Tn (−x) = (−1)n Tn (x).
(A-4.27)
The Rodrigues formula is √ Tn (x) =
1 π(−1)n (1 − x2 ) 2 dn 2 n− 2 1 − x . · 1 dxn 2n (n − 2 )! 1
(A-4.28)
The orthogonal relation for Tn (x) is
1 −1
1 2 −2
1−x
⎧ ⎪ ⎨0 Tm (x) Tn (x) dx =
⎪ ⎩
if m = n,
π 2
if m = n,
π
if m = n = 0.
(A-4.29)
The Chebyshev polynomials of the second kind, Un (x), are defined by − 1 Un (x) = 1 − x2 2 sin (n + 1) cos−1 x ,
−1 ≤ x ≤ 1.
(A-4.30)
The generating function for Un (x) is
1 − 2x t + t2
−1
=
∞
Un (x)tn ,
|x| < 1, |t| < 1.
n=0
The first seven Chebyshev polynomials Un (x) are given by U0 (x) = 1, U1 (x) = 2x, U2 (x) = 4x2 − 1,
(A-4.31)
710
A Some Special Functions and Their Properties
U3 (x) U4 (x) U5 (x) U6 (x)
= 8x3 − 4x, = 16x4 − 12x2 + 1, = 32x5 − 32x3 + 6x, = 64x6 − 80x4 + 24x2 − 1.
The differential equation for y = Un (x) is 1 − x2 y − 3xy + n(n + 2)y = 0.
(A-4.32)
The recurrence relations are
Un+1 (x) − 2x Un (x) + Un−1 (x) = 0, 1 − x Un (x) + nxUn (x) − (n + 1)Un−1 (x) = 0. 2
(A-4.33) (A-4.34)
The parity relation is Un (−x) = (−1)n Un (x).
(A-4.35)
The Rodrigues formula is √ n+ 12 π(−1)n (n + 1) dn Un (x) = 1 − x2 . 1 n 1 n+1 2 dx 2 2 (n + 2 )!(1 − x )
(A-4.36)
The orthogonal relation for Un (x) is
1 −1
1 − x2
12
Um (x) Un (x) dx =
π δmn . 2
(A-4.37)
A-5 Laguerre and Associated Laguerre Functions The Laguerre polynomials Ln (x) are defined by the Rodrigues formula Ln (x) = ex
dn n −x x e , dxn
where n = 0, 1, 2, 3, . . . . The first seven Laguerre polynomials are L0 (x) = 1, L1 (x) = 1 − x, L2 (x) = 2 − 4x + x2 , L3 (x) = 6 − 18x + 9x2 − x3 , L4 (x) = 24 − 96x + 72x2 − 16x3 + x4 , L5 (x) = 120 − 600x + 600x2 − 200x3 + 25x4 − x5 , L6 (x) = 720 − 4320x + 5400x2 − 2400x3 + 450x4 − 36x5 + x6 .
(A-5.1)
A-5 Laguerre and Associated Laguerre Functions
711
The generating function is
∞ xt = tn Ln (x). (1 − t)−1 exp 1−t n=0
(A-5.2)
Ln (0) = 1.
(A-5.3)
In particular, The orthogonal relation for the Laguerre polynomial is ∞ e−x Lm (x)Ln (x) dx = (n!)2 δnm .
(A-5.4)
0
The recurrence relations are (n + 1)Ln+1 (x) = (2n + 1 − x)Ln (x) − nLn−1 (x), xLn (x) = nLn (x) − nLn−1 (x), Ln (x) = Ln−1 (x) − Ln−1 (x).
(A-5.5) (A-5.6) (A-5.7)
The Laguerre polynomials, y = Ln (x), satisfy the Laguerre differential equation xy + (1 − x)y + ny = 0.
(A-5.8)
The associated Laguerre polynomials are defined by dm Ln (x) for n ≥ m. dxm The generating function for Lm n (x) is ∞ xz n = Lm |z| < 1. (1 − z)−(m+1) exp − n (x)z , 1−z n=0 Lm n (x) =
(A-5.9)
(A-5.10)
It follows from this that
(n + m)! . n!m! The associated Laguerre function satisfies the recurrence relation Lm n (0) =
m m (n + 1)Lm n+1 (x) = (2n + m + 1 − x)Ln (x) − (n + m)Ln−1 (x), d m (x) = n Lm x Lm n (x) − (n + m)Ln−1 (x). dx n The associated Laguerre function, y = Lm n (x), satisfies the associated differential equation xy + (m + 1 − x)y + ny = 0.
The Rodrigues formula for
Lm n (x)
(A-5.12) (A-5.13) Laguerre (A-5.14)
is
ex x−m dn −x n+m e x . n! dxn The orthogonal relation for Lm n (x) is ∞ (n + m)! m δn l . e−x xm Lm n (x)Ll (x) dx = n! 0 Lm n (x) =
(A-5.11)
(A-5.15)
(A-5.16)
712
A Some Special Functions and Their Properties
A-6 Hermite Polynomials and Weber–Hermite Functions The Hermite polynomials Hn (x) are defined by the Rodrigues formula dn exp −x2 , Hn (x) = (−1)n exp x2 dxn
(A-6.1)
where n = 0, 1, 2, 3, . . . . The first seven Hermite polynomials are H0 (x) = 1, H1 (x) = 2x, H2 (x) = 4x2 − 2, H3 (x) = 8x3 − 12x, H4 (x) = 16x4 − 48x2 + 12, H5 (x) = 32x5 − 16x3 + 120x, H6 (x) = 64x6 − 480x4 + 720x2 − 120. The generating function is ∞ n t exp 2xt − t2 = Hn (x). n! n=0
(A-6.2)
It follows from (A-6.2) that Hn (x) satisfies the parity relation Hn (−x) = (−1)n Hn (x).
(A-6.3)
Also, it follows from (A-6.2) that H2n+1 (0) = 0,
H2n (0) = (−1)n
(2n)! . n!
(A-6.4)
The recurrence relations for Hermite polynomials are Hn+1 (x) − 2xHn (x) + 2nHn−1 (x) = 0, Hn (x) = 2xHn−1 (x).
(A-6.5) (A-6.6)
The Hermite polynomials, y = Hn (x), are solutions of the Hermite differential equation (A-6.7) y − 2xy + 2ny = 0. The orthogonal property of Hermite polynomials is ∞ √ exp −x2 Hn (x)Hm (x) dx = 2n n! πδmn . −∞
With repeated use of integration by parts, it follows from (A-6.1) that
(A-6.8)
A-7 Mittag-Leffler Function
∞
−∞ ∞ −∞
exp −x2 Hn (x)xm dx = 0,
m = 0, 1, . . . , (n − 1),
√ exp −x2 Hn (x)xn dx = πn!.
(A-6.9) (A-6.10)
The Weber–Hermite function, or simply Hermite function, 2 x y = hn (x) = exp − Hn (x) 2 satisfies the Hermite differential equation y + λ − x2 y = 0,
713
x ∈ R,
(A-6.11)
(A-6.12)
where λ = 2n + 1. If λ = 2n + 1, then y is not finite as |x| → ∞. The Hermite functions {hn (x)}∞ 0 form an orthogonal basis for the Hilbert space 2 L (R) with weight function 1. They satisfy the following fundamental properties: hn (x) + xhn (x) − 2n hn−1 (x) = 0, hn (x) − xhn (x) + hn+1 (x) = 0, hn (x) − x2 hn (x) + (2n + 1)hx = 0, ˜ n (k) = (−i)n hn (k). F hn (x) = h The normalized Weber–Hermite functions are given by 2 x −n/2 − 14 − 12 Hn (x). π (n!) exp − ψn (x) = 2 2
(A-6.13)
Physically, they represent quantum-mechanical oscillator wave functions. The graphs of these functions are shown in Figure A.10.
A-7 Mittag-Leffler Function Another important function that has widespread use in fractional calculus and fractional differential equation is the Mittag-Leffler function. The Mittag-Leffler function is an entire function defined by the series Eα (z) =
∞
zn , Γ (αn + 1) n=0
α > 0.
(A-7.1)
The graph of the Mittag-Leffler function is shown in Figure A.11. The generalized Mittag-Leffler function, Eα,β (z), is defined by Eα,β (z) =
∞
zn , Γ (αn + β) n=0
α, β > 0.
(A-7.2)
714
A Some Special Functions and Their Properties
Fig. A.10 The normalized Weber–Hermite functions.
Fig. A.11 Graph of the Mittag-Leffler function Eα (x).
Also the inverse Laplace transform yields
m!sα−β (m) −1 L = tαm+β−1 Eα,β ±atα , α m+1 ¯ (s +a)
(A-7.3)
where (m)
Eα,β (z) =
dm Eα,β (z). dz m
(A-7.4)
A-8 The Jacobi Elliptic Integrals and Elliptic Functions
715
Obviously, E1,1 (z) = E1 (z) = ez .
Eα,1 (z) = Eα (z),
(A-7.5)
A-8 The Jacobi Elliptic Integrals and Elliptic Functions The parametric equation of an ellipse is given by x = a sin θ,
(a > b), 0 ≤ θ ≤ φ.
y = b cos θ,
(A-8.1)
Using the arclength formula from calculus, the length of the elliptic arc (A-8.1) is ds2 = dx2 + dy2 = a2 cos2 θ + b2 sin2 θ dθ2 a2 − b2 2 = a2 1 − sin θ dθ2 = a2 1 − m2 sin2 θ dθ2 , (A-8.2) 2 a ) 2 < 1 is the eccentricity of the ellipse. where e = m = ( a a−b 2 Consequently, (A-8.2) gives the length of the elliptic arc 2
2
1
s
φ
ds = a
s= 0
1 − m2 sin2 θ dθ.
(A-8.3)
0
This integral cannot be evaluated in terms of elementary functions. Because of its origin, it is called an elliptic integral. In general, there are three classes of elliptic integrals, called elliptic integrals of the first, second and third kinds, and they defined by
φ
F (φ, m) =
0
φ
E(φ, m) =
dθ 1 − m2 sin2 θ
(A-8.4)
,
1 − m2 sin2 θ dθ,
(A-8.5)
0
φ
Π(φ, m, n) = 0
(1 +
n2
dθ , sin θ) 1 − m2 sin2 θ 2
(m = n),
(A-8.6)
where the parameter φ is called the amplitude, φ = am(F, m) so that am(0, m) = 0 and m (0 < m < 1) is called the modulus. When φ = π2 , (A-8.4)–(A-8.6) are referred to as complete elliptic integrals of the first, second and third kinds, and they are denoted by special symbols: K(m) = F ( π2 , m), E(m) = E( π2 , m) and Π(m, n) = Π( π2 , m, n). Putting x = sin θ, 0 ≤ θ ≤ φ, the first, second and third elliptic integrals can be written in equivalent forms as F (φ, m) = 0
sin φ
dx , (1 − x2 )(1 − m2 x2 )
(A-8.7)
716
A Some Special Functions and Their Properties
sin φ
E(φ, m) = 0
sin φ
Π(φ, m, n) = 0
√ 1 − m2 x2 √ dx, 1 − x2 dx . 2 2 (1 + n x ) (1 − x2 )(1 − m2 x2 )
(A-8.8) (A-8.9)
Using (A-8.7)–(A-8.9), the Jacobi elliptic functions, sn(u, m), cn(u, m) and dn(u, m), are defined by sn(u, m) = sin φ,
(A-8.10)
12
cn(u, m) = cos φ = 1 − sn u , 1 1 dn(u, m) = 1 − m2 sin2 φ 2 = 1 − m2 sn 2 u 2 2
(A-8.11) (A-8.12)
so that sn(−u) = − sn u, sn(0) = 0, and
cn(−u) = cn u, cn(0) = dn(0) = 1.
dn(−u) = dn u,
(A-8.13) (A-8.14)
The following limiting results also hold lim sn(u, m) = sin u,
m→0
lim cn(z, m) = cos u,
m→0
(A-8.15)
lim dn(z, m) = 1,
m→0
lim sn(u, m) = tanh u,
m→1
lim cn(u, m) = sech u,
m→1
(A-8.16)
lim dn(u, m) = sech u.
m→1
Making reference to Dutta and Debnath (1965) without proof, we state the following basic properties of the Jacobi elliptic functions: sn 2 u + cn 2 u = 1,
dn 2 u + m2 sn 2 u = 1,
d sn u = cn u dn u, du
dn 2 u − m2 cn 2 u = 1 − m2 , (A-8.17)
d cn u = − sn u dn u, du
(A-8.18)
and d dn u = −m2 sn u cn u. du
(A-8.19)
Putting sn(u, m) = x in the first result in (A-8.18) gives the differential equation dx = 1 − x2 1 − m2 x2 , (A-8.20) du so that it leads to the Legendre normal form for the sn-function sn(u,m) dx . u= 2 )(1 − m2 x2 ) (1 − x 0
(A-8.21)
A-8 The Jacobi Elliptic Integrals and Elliptic Functions
717
Similarly, it follows from (A-8.18)–(A-8.19) that the Legendre normal forms for cn and dn functions are 1 dx , (A-8.22) u= 2 (1 − x )(m2 + m2 x2 ) cn(u,m) 1 dx u= , (A-8.23) 2 )(x2 − m2 ) (1 − x dn(u,m) √ where m = 1 − m2 is called the complementary modulus of the elliptic integral. The complete elliptic integrals are then given by π π , m = K m , , m = E m . (A-8.24) E(m) = E K(m) = F 2 2 The limiting values of K(m) and E(m) are given as follows: π , m→0 2 lim K(m) = K(1) = ∞, lim K(m) = K(0) =
m→1
π , m→0 2 lim E(m) = E(1) = 1. lim E(m) = E(0) =
m→1
(A-8.25) (A-8.26)
Finally, the addition theorems for sn, cn, and dn functions are sn u cn v dn v + sn v cn u dn u , (1 − m2 sn 2 u sn 2 v) cn u cn v − sn u dn u sn v dn v cn(u + v) = , (1 − m2 sn 2 u sn 2 v) dn u dn v − m2 sn u cn u sn v cn v , dn(u + v) = (1 − m2 sn 2 u sn 2 v) sn(u + v) =
(A-8.27) (A-8.28) (A-8.29)
where sn(u + 4K) = sn u, cn(u + 4K) = cn u, and dn(u + 2K) = dn u so that sn and cn are periodic functions of period 4K, and dn is a periodic function of period 2K.
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
The main purpose of this appendix is to discuss Fourier series and generalized functions, and to state their basic properties that are most frequently used in the theory and applications of ordinary and partial differential equations. The subject is, of course, too vast to be treated adequately in so short a space, so that only the more important results will be stated. Included are basic properties of Fourier and Laplace transforms which are used in finding solutions of ordinary and partial differential equations. For a fuller discussion of these topics and of further properties of these functions the reader is referred to the standard treatises on the subjects including Debnath and Bhatta (2007).
B-1 Fourier Series and Its Basic Properties If f (x) is a periodic function of period 2π defined in (−π, π), then f (x) can be represented as an infinite series in terms of trigonometric functions in the form f (x) =
∞ 1 (an cos nx + bn sin nx). a0 + 2 n=1
(B-1.1)
This is known as the Fourier Series. If we assume that the infinite series is term-byterm integrable on (−π, π), then π π ∞ 1 a0 + f (x) dx = (an cos nx + bn sin nx) dx = πa0 −π −π 2 n=1 so that a0 =
1 π
π
f (x) dx.
(B-1.2)
−π
Multiplying both sides of (B-1.1) by cos mx and integrating the resulting series from −π to π gives L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1,
720
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms π
f (x) cos mx dx π ∞ 1 (an cos nx + bn sin nx) cos mx dx = πan , = a0 + −π 2 n=1
−π
m = n.
Thus, an =
1 π
π
f (x) cos nx dx.
(B-1.3)
−π
Similarly, multiplying (B-1.1) by sin mx and integrating over (−π, π) gives 1 π bn = f (x) sin nx dx. (B-1.4) π −π The Fourier coefficients an and bn given by (B-1.2)–(B-1.4) are known as the Euler– Fourier formulas. If f (x) is an even function of x defined on [−π, π], then 1 π an = f (x) cos nx dx π −π π 2 = f (x) cos nx dx, n = 0, 1, 2, . . . , (B-1.5) π 0 π 1 f (x) sin nx dx = 0, n = 1, 2, 3, . . . . (B-1.6) bn = π −π Hence, the Fourier series of an even function f (x) can be written as f (x) =
∞ 1 an cos nx, a0 + 2 n=1
(B-1.7)
where an are given by (B-1.5). Similarly, if f (x) is an odd function of x on (−π, π), the Fourier series of an odd function is given by f (x) =
∞
bn sin nx,
(B-1.8)
n=1
where an = 0 for all n, and bn are given by 1 π bn = f (x) sin nx dx π −π 2 π f (x) sin nx dx, = π 0
n = 1, 2, 3, . . . .
(B-1.9)
B-1 Fourier Series and Its Basic Properties
721
Fig. B.1 The function f (x) and its extension.
Fig. B.2 Graph of f (x) = x2 and its extension.
Example B-1.1. The Fourier series of f (x) (see Figure B.1) 0 if − π < x < 0, f (x) = x if 0 ≤ x < π is given by (B-1.1), where 1 π 1 π π a0 = f (x) dx = x dx = , π −π π 0 2 1 π 1 π 1 an = (−1)n − 1 , f (x) cos nx dx = x cos nx dx = 2 π −π π 0 πn 1 π 1 bn = x sin nx dx = − (−1)n . π 0 n Hence, the Fourier series of f (x) is f (x) =
∞ 2 (−1)n π − sin nx . cos(2n − 1)x + 4 n=1 π(2n − 1)2 n
(B-1.10)
Example B-1.2. The Fourier series of the (see Figure B.2) function f (x) = x2 ,
−π ≤ x ≤ π with f (x ± 2nπ) = f (x), n = 1, 2, 3, . . . ,
722
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
Fig. B.3 The triangular wave function and its extension.
is x2 =
∞ (−1)n π2 +4 cos nx. 3 n2 n=1
(B-1.11)
Since f (x) is even, bn = 0 for all n ≥ 1, it turns out that 1 π 2 2 π 2 2π 2 a0 = , x dx = x dx = π −π π 0 3 1 π 2 2 π 2 an = x cos nx dx = x cos nx dx π −π π 0 4 = 2 (−1)n , n ≥ 1. n Example B-1.3. The triangular wave function (see Figure B.3) −x if − π ≤ x < 0, f (x) = |x| = x if 0 ≤ x < π, with f (x ± 2πn) = f (x) has the Fourier cosine series representation f (x) =
∞ 4 π 1 − cos(2n − 1)x. 2 π n=1 (2n − 1)2
In this case, f (x) is even, thus, bn = 0 for all n ≥ 1, and 1 π 2 π |x| dx = x dx = π, a0 = π −π π 0 1 π 2 π an = |x| cos nx dx = x cos nx dx π −π π 0 2 = (−1)n − 1 , 2 πn and so
an =
0
if n is even,
− πn4 2
if n is odd.
(B-1.12)
B-1 Fourier Series and Its Basic Properties
723
Fig. B.4 The sawtooth wave function.
Example B-1.4. The sawtooth wave function (see Figure B.4) −π < x < π,
f (x) = x,
with f (x) = f (x ± 2nπ), n = 1, 2, 3, . . . , has the Fourier sine series expansion f (x) = 2
∞
(−1)n+1
n=1
sin nx . n
(B-1.13)
In this case, f (x) is odd and hence, an = 0 for all n ≥ 0, and 1 π 2 π 2 x sin nx dx = x sin nx dx = (−1)n+1 . bn = π −π π 0 n Example B-1.5. The Fourier series representation of the square wave function (see Figure B.5) defined by ⎧ 1 if 0 < x ≤ π, ⎪ ⎪ ⎨ if x = 0, f (x) = sgn(x) = 0 ⎪ ⎪ ⎩ −1 if − π ≤ x < 0 is given by f (x) =
∞ 4 1 sin(2n − 1)x. π n=1 (2n − 1)
(B-1.14)
Obviously, f (x) is odd, and hence, an = 0 for all n ≥ 0, and bn is given by 1 π 2 π bn = sgn x sin nx dx = sin nx dx π −π π 0 0 if n is even, 2 1 − (−1)n = = 4 π n if n is odd. nπ
724
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
Fig. B.5 The square wave function and its extension.
Fig. B.6 The triangular wave function and its extension.
Example B-1.6. The triangular wave function (see Figure B.6) f (x) on [−π, π] is given by π + x if − π ≤ x ≤ 0, f (x) = π − x if 0 ≤ x ≤ π. Since f (x) is even, bn = 0 for n ≥ 1, and 1 π 2 π a0 = f (x) dx = (π − x) dx = π, π −π π 0 π 2 2 1 n 1 − (−1) (π − x) cos nx dx = an = π 0 π n2 0 if n is even, 2 = · πn2 2 if n is odd. Thus, the Fourier series for f (x) is ∞ a0 + f (x) = an cos nx 2 n=1 ∞ 1 4 π cos(2n − 1)x. = + 2 π n=1 (2n − 1)2
(B-1.15)
B-1 Fourier Series and Its Basic Properties
725
If f (x) is a periodic function of period 2l and is defined on [−l, l], then the Fourier representation of f (x) is f (x) =
∞ nπx nπx a0 + bn sin , + an cos 2 l l n=1
(B-1.16)
where a0 , an , and bn are given by the Euler–Fourier formulas: a0 =
1 l
l
(B-1.17)
f (x) dx, −l l
nπx dx, f (x) cos l −l 1 l nπx bn = f (x) sin dx. l −l l
an =
1 l
(B-1.18) (B-1.19)
If f (x) is an even function of period 2l defined on [−l, l], then f (x) =
∞ a0 nπx , + an cos 2 l n=1
where bn = 0, n ≥ 1, and 2 l nπx dx, an = f (x) cos l 0 l
n = 0, 1, 2, . . . .
(B-1.20)
(B-1.21)
If f (x) is an odd function of period 2l, then f (x) =
nπx bn sin , l n=1 ∞
(B-1.22)
where an = 0, n ≥ 0, and 2 bn = l
0
l
nπx f (x) sin l
dx.
(B-1.23)
The functions f (x) in all Examples B-1.1–B-1.6 can be defined on [−l, l] and the corresponding Fourier series can be obtained directly by calculating Fourier coefficients an and bn , or by using the transformation x = πt l . In Example B-1.1, the Fourier series on [−l, l] is ∞ 2 l nπx (2n − 1)πx (−1)n l . + sin cos f (x) = − 4 π n=1 π(2n − 1)2 l n l (B-1.24) In Example B-1.2, the Fourier series on [−l, l] is given by
726
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
∞ l2 nπx 4l2 (−1)n f (x) = cos . + 2 3 π n=1 n2 l In Example B-1.3, the Fourier series on [−l, l] is
∞ 1 l 4l xπ f (x) = − . cos (2n − 1) 2 π 2 n=1 (2n − 1)2 l When l = π, this reduces to (B-1.12). In Example B-1.4, the Fourier series on [−l, l] is ∞ nπx 2l (−1)n+1 . sin f (x) = π n=1 n l In Example B-1.5, the Fourier series of sgn(x) on [−l, l] is
∞ 1 4 xπ sgn(x) = . sin (2n − 1) π n=1 (2n − 1) l
(B-1.25)
(B-1.26)
(B-1.27)
(B-1.28)
In Example B-1.6, the Fourier series on [−l, l] is f (x) =
∞ 1 4l l πx + 2 cos(2n − 1) . 2 π n=1 (2n − 1)2 l
(B-1.29)
It is sometimes convenient to represent a function f (x) by a Fourier series in complex form. This expansion can easily be derived from the Fourier series (B-1.1), that is, ∞
a0 + (an cos nx + bn sin nx) 2 n=1 ∞ bn inx a0 1 inx + an e + e−inx + e = − e−inx 2 2 2i n=1 ∞ 1 1 a0 inx −inx + (an + ibn )e = + (an − ibn )e 2 2 2 n=1
f (x) =
= c0 +
∞ ∞ inx cn e + c−n e−inx = cn einx , n=1
where
n=−∞
π a0 1 f (x) dx, = 2 2π −π π 1 1 f (x)(cos nx − i sin nx) dx cn = (an − ibn ) = 2 2π −π π 1 = f (x)e−inx dx, 2π −π c0 =
B-1 Fourier Series and Its Basic Properties
727
π 1 1 (an + ibn ) = f (x)(cos nx + i sin nx) dx 2 2π −π π 1 f (x)einx dx = cn . = 2π −π
c−n =
Thus, we obtain the Fourier series of f (x) in complex form f (x) =
∞
cn einx ,
−π < x < π,
(B-1.30)
n=−∞
where the Fourier coefficients cn are given by π 1 f (x)e−inx dx. cn = 2π −π
(B-1.31)
1 Multiplying (B-1.30) by 2π f (x) and integrating from −π to π gives π π ∞ 1 1 f 2 (x) dx = cn f (x)einx dx 2π −π 2π −π n=−∞ ∞ cn · c−n =
=
n=−∞ ∞
∞
cn cn =
n=−∞
|cn |2 .
(B-1.32)
n=−∞
Thus, (B-1.32) is known as the Parseval formula for a complex Fourier series. We next consider the nth partial sum a0 (ak cos kx + bk sin kx), + 2 n
sn (x) =
(B-1.33)
k=1
π of the Fourier series for f (x) in (B-1.1) defined on [−π, π]. If −π f 2 (x) dx exists and is finite, then π 2 0≤ f (x) − sn (x) dx π π −π π f 2 (x) dx − 2 f (x)sn (x) dx + s2n (x) dx. (B-1.34) = −π
−π
−π
It follows from the definition of the Fourier coefficients (B-1.2)–(B-1.4) and the orthogonality of the cosine and sine functions that π π n a0 + f (x)sn (x) dx = f (x) (ak cos kx + bk sin kx) dx 2 −π −π k=1
πa20 +π a2k + b2k . = 2 n
k=1
728
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
Similarly, it turns out that π
−π
π
s2n (x) dx
=
−π π
= −π
=
πa20 2
2 n a0 (ak cos kx + bk sin kx) dx + 2 k=1 π π n a20 cos2 kx dx + b2k sin2 kx dx a2k dx + 4 −π −π k=1
n 2 +π ak + b2k . k=1
Consequently, (B-1.34) reduces to π 2 0≤ f (x) − sn (x) dx −π π n 2 πa20 2 2 f (x) dx − = +π ak + bk . 2 −π
(B-1.35)
k=1
This leads to the inequality
1 a20 2 + ak + b2k ≤ 2 π n
π
f 2 (x) dx,
(B-1.36)
−π
k=1
and since the right-hand side of (B-1.36) is independent of n, it follows in the limit as n → ∞ that ∞ 1 π 2 a20 2 2 + ak + bk ≤ f (x) dx. (B-1.37) 2 π −π k=1
This is known as the Bessel inequality for a Fourier series. Since the left-hand side of (B-1.36) is non-decreasing and is bounded above, the series ∞
a20 2 + ak + b2k 2
(B-1.38)
k=1
converges. Thus, the necessary condition for the convergence of the series (B-1.38) is that lim ak = 0 and
k→∞
lim bk = 0.
k→∞
(B-1.39)
That is, k→∞
π
lim
f (x) cos kx dx = 0, −π
π
lim
k→∞
f (x) sin kx dx = 0. −π
These results are known as the Riemann–Lebesgue Lemma.
(B-1.40)
B-1 Fourier Series and Its Basic Properties
The Fourier series is said to converge in the mean to f (x) when π 2 f (x) − sn (x) dx = 0. lim n→∞
729
(B-1.41)
−π
If the Fourier series converges in the mean to f (x), then ∞
a20 2 1 + an + b2n = 2 π n=1
π
f 2 (x) dx.
(B-1.42)
−π
This is called Parseval’s relation, and it is one of the fundamental results in the theory of Fourier series. This relation can formally be derived from the convergence of the Fourier series to f (x) on [−π, π]. In other words, if f (x) =
∞ 1 (an cos nx + bn sin nx), a0 + 2 n=1
(B-1.43)
where a0 , an , and bn are given by (B-1.2)–(B-1.4), we multiply by (B-1.43) by 1 f (x) and integrate the resulting expression from −π to π to obtain π 1 π 2 f (x) dx π −π a0 π = f (x) dx 2π −π π π ∞ 1 1 + an f (x) cos nx dx + bn f (x) sin nx dx . (B-1.44) π π −π −π n=1 Replacing all integrals on the right hand side of (B-1.44) by the Fourier coefficients gives the Parseval relation (B-1.42). If two (2π)-periodic integrable functions f (x) and g(x) defined on [−π, π] have the Fourier series expansions f (x) =
∞ 1 a0 + (an cos nx + bn sin nx), 2 n=1
(B-1.45)
g(x) =
∞ 1 (αn cos nx + βn sin nx), α0 + 2 n=1
(B-1.46)
where a0 , an , and bn are given by (B-1.1)–(B-1.3), and α0 , αn , and βn are given by results similar to those of (B-1.1)–(B-1.3), then the following Parseval’s relations hold ∞ 2 1 π 1 f (x) + g(x) dx = (a0 + α0 )2 + (an + αn )2 + (bn + βn )2 , π −π 2 n=1 (B-1.47)
730
1 π
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
π −π
f (x) − g(x)
2
∞ 1 2 dx = (a0 − α0 ) + (an − αn )2 + (bn − βn )2 . 2 n=1
(B-1.48) Subtracting (B-1.48) from (B-1.47) yields 1 π
π
f (x)g(x) dx = −π
∞ 1 (an αn + bn βn ). a0 α0 + 2 n=1
(B-1.49)
This is a general Parseval relation for the product function f (x)g(x). When f = g, (B-1.49) reduces to (B-1.42). In the context of the complex Fourier series expansion (B-1.30) of a (2π)periodic function f (x), we can replace the Fourier coefficient cn by f(n) so that π 1 e−inx f (x) dx. (B-1.50) cn = f(n) = 2π −π The concept of convolution of two (2π)-periodic integrable functions f and g on R arises naturally, so that we define their convolution (f ∗ g)(x) on [−π, π] by π 1 (f ∗ g)(x) = f (x − t)g(t) dt. (B-1.51) 2π −π Physically, the convolution (f ∗ g)(x) represents an integral output of the two functions f and g in contrast to the ordinary pointwise product (output) f (x)g(x). Clearly, the convolution is commutative, that is, π 1 f (ξ)g(x − ξ) dξ = (g ∗ f )(x). (B-1.52) (f ∗ g)(x) = 2π −π Another interpretation of a convolution is that it represents an average (or mean) value. In particular, if g = 1 in (B-1.52), then f ∗ g is constant and is equal to π 1 f (ξ) dξ. (B-1.53) (f ∗ 1)(x) = 2π −π This means that (f ∗ 1)(x) represents the average value of f (x) on [−π, π]. In addition to commutativity, the convolution satisfies the following algebraic and analytic properties for any constant α and β: f ∗ (αg + βh) = α(f ∗ g) + β(f ∗ h) (Distributive),
(B-1.54)
(f ∗ g) ∗ h = f ∗ (g ∗ h) (Associative),
(B-1.55)
(f ∗ g)(x) is continuous
(B-1.56)
f ∗ g(n) = f(n) g (n)
(Continuity),
(Convolution).
(B-1.57)
B-1 Fourier Series and Its Basic Properties
To prove (B-1.57), we use the definition π 1 (f ∗ g)(x)e−inx dx f ∗ g(n) = 2π −π π π 1 1 e−inx f (t)g(x − t) dt dx = 2π −π 2π −π π π 1 1 = e−int f (t) e−in(x−t) g(x − t) dx dt, 2π −π 2π −π π π 1 1 −int −ins e f (t) dt g(s)e ds = 2π −π 2π −π = f(n) g (n). The nth partial sum of a complex Fourier series (B-1.30) is π n n 1 sn (x) = ck eikx = eikx f (t)e−ikt dt 2π −π k=−n k=−n n π 1 ik(x−t) dt f (t) e = 2π −π k=−n π 1 = f (t)Dn (x − t) dt, 2π −π = Dn (x) ∗ f (x),
731
x−t=s
(B-1.58) (B-1.59)
where Dn (x) is called the Dirichlet kernel defined by Dn (x) =
n
eikx = 1 +
k=−n
= 1+
n
n
eikx + e−ikx
k=1
2 cos kx.
(B-1.60)
k=1
We next make the following observations regarding the genesis of the convolution in the theory of Fourier. The convolution (Dn ∗f )(x) arises in the nth partial sum sn (x) of the Fourier series of f (x). Thus, the problem of understanding sn (x) reduces to that of (Dn ∗ f )(x). It follows from (B-1.60) that x 1 1 x x Dn (x) sin = sin + sin (cos x + cos 2x + · · · + cos nx) 2 2 2 2 2 3x 5x x 1 x 3x 1 sin + sin − sin − sin = sin + 2 2 2 2 2 2 2 1 1 x − sin n − x + · · · + sin n + 2 2 1 1 = sin n + x. 2 2
732
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
Fig. B.7 Graph of Dn (x = θ) against (x = θ).
Thus, the exact form of the Dirichlet kernel is Dn (x) =
sin(n + 12 )x . sin x2
(B-1.61)
This reveals that the denominator of the Dirichlet kernel Dn (x) vanishes at the points x = 2πm which are removable points of discontinuity. Furthermore, it follows from (B-1.60) that Dn (x) is an even function with period 2π and satisfies π Dn (x) dx = 2π. (B-1.62) −π
The graph of Dn (x = θ) is shown in Figure B.7. It looks very similar to that of the diffusion kernel as shown in Figure 1.6 in Chapter 1 except for its symmetric oscillatory trail in both positive and negative θ-axes. It follows from (B-1.58), by putting t − x = ξ and noting that Dn (ξ) is even, that π−x 1 f (x + ξ)Dn (ξ) dξ sn (x) = 2π −π−x π 1 = f (x + ξ)Dn (ξ) dξ. (B-1.63) 2π −π We close this section by stating the Pointwise Convergence Theorem: If f (x) is a piecewise smooth and periodic function with period 2π on [−π, π], then, for any x in (−π, π), n a0 + lim sn = lim (ak cos kx + bk sin kx) n→∞ n→∞ 2 k=1
1 = f (x + 0) + f (x − 0) , 2
(B-1.64)
B-1 Fourier Series and Its Basic Properties
733
or equivalently, ∞ 1 a0 + (ak cos kx + bk sin kx) = f (x+) + f (x−) , 2 2
(B-1.65)
k=1
where
1 π f (x) cos kx dx, k = 0, 1, 2, 3, . . . , π −π 1 π bk = f (x) sin kxdx, k = 1, 2, 3, . . . . π −π
ak =
(B-1.66) (B-1.67)
We refera to Myint-U and Debnath (2007) for its proof. Another remarkable feature of Fourier series of a function at its ordinary points of discontinuity deals with the behavior of its nth partial sums sn (x) as n → ∞. At points where f (x) is continuous, the nth partial sums sn (x) approach smoothly the value f (x) as n → ∞. However, for the functions f (x) = x in Example B-1.4 or f (x) = sgnx in Example B-1.5, the graphs of their partial sums exhibit a large error in the neighborhood of points of discontinuity at x = 0 and x = ±π independent of the number of terms in their partial sums. In other words, the partial sums do not converge smoothly to the mean value. Instead, they overshoot the mark at each end of the jumps of the function. The explanation of this phenomenon was first provided by J.W. Gibbs (1839–1903) in 1899, who showed that overshooting was not the result of computational errors. This feature is typical for the Fourier series of a function at the points of discontinuity, and is now universally known as the Gibbs phenomenon. One of the most effective and useful applications of Fourier series deals with the summation of infinite series in closed form which is one of the major problems in mathematics. We shall use Examples B-1.1–B-1.6 to derive the sums of many important numerical series. Substituting x = 0 in (B-1.10) gives the numerical series 0 = f (0) =
∞ 1 2 π − . 4 π n=1 (2n − 1)2
Hence, ∞
1 π2 , = 2 (2n − 1) 8 n=1 or equivalently, 1 1 1 1 π2 . + + + + · · · = 12 32 52 72 8
(B-1.68)
This can be used to obtain another numerical series S=
1 1 1 1 π2 1 . + 2 + 2 + 2 + ··· + 2 + ··· = 2 1 2 3 4 n 6
(B-1.69)
734
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
In fact,
1 1 1 S= + 2 + 2 + ··· 2 1 3 5 1 π2 + S. = 8 4
1 1 1 1 1 + + 2 + 2 + 2 + ··· 4 12 2 3 4
2
Thus, S(1 − 14 ) = π8 , which gives (B-1.69). Putting x = 0 in (B-1.11) leads to the numerical series 0=
∞ (−1)n π2 , +4 3 n2 n=1
or equivalently, ∞ π2 (−1)n+1 = . n2 12 n=1
Thus, 1 1 1 1 π2 . − + − + · · · = 12 22 32 42 12 Substituting x =
π 2
(B-1.70)
in the series (B-1.13) gives
∞ (−1)n+1 π nπ =2 sin 2 2 n n=1 π sin 2π sin 3π sin 4π sin 2 2 2 2 − + − + ··· =2 1 2 3 4 1 1 1 = 2 1 − + − + ··· . 3 5 7
Therefore, 1−
1 1 1 π + − + ··· = . 3 5 7 4
(B-1.71)
This is celebrated Leibniz series for π discovered by Leibniz in 1673. It is also known as the Gregory series independently discovered by James Gregory (1638–1675) in around 1670. Putting x = π4 in (B-1.13) gives another numerical series π 1 1 1 1 1 1 1 1 1 1 1 − + − + ··· . = √ 1+ − − + + − ··· − 8 3 5 7 9 11 2 3 5 7 2 In view of (B-1.71), this leads to the following series
B-1 Fourier Series and Its Basic Properties
1+
1 π 1 1 1 1 − − + + − ··· = √ . 3 5 7 9 11 2 2
735
(B-1.72)
Subtracting (B-1.70) from (B-1.69) yields 1 1 1 π2 1 + + + + · · · = . 22 42 62 82 24
(B-1.73)
In Example B-1.2, the Fourier series for f (x) = x2 is given by (B-1.11). It follows from the Parseval relation (B-1.42) that ∞ ∞ 1 π 4 2π 4 2 2π 4 16 + + x dx = an = , π −π 9 9 n4 n=1 n=1 or equivalently, ∞
2π4 16 2π 4 = + . 5 9 n4 n=1 Thus, ∞ 1 π4 . = 4 n 90 n=1
(B-1.74)
If we apply the Parseval formula (B-1.42) to Example B-1.3, we can derive the following numerical series ∞
π4 1 . = 4 (2n − 1) 96 n=1
(B-1.75)
A similar calculation can be used for the Fourier series of f (x) = x(π − x), 0 < x < π, in the form 8 sin x sin 3x sin 5x f (x) = + + + · · · . (B-1.76) π 13 33 53 Therefore, we can show that 1 1 1 π3 1 − 3 + 3 − 3 + ··· = , 3 1 3 5 7 32 ∞ π6 1 . = n6 945 n=1
(B-1.77) (B-1.78)
In 1.15 Exercises, Problem 10, the initial conditions are f (x) and g(x) = 0 defined on 0 ≤ x ≤ l by 2hx if 0 ≤ x ≤ 2l , l f (x) = (B-1.79) 2h(l−x) if 2l ≤ x ≤ l, l
736
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
where the midpoint of the string of length l is held at a vertical distance h from the equilibrium position for t < 0 and released at time t = 0. We expand f (x) in a Fourier sine series (an ≡ 0) so that f (x) =
∞ n=1
bn sin
nπx , l
(B-1.80)
where nπx dx f (x) sin l 0 l/2 l 2 2h(l − x) nπx 2hx nπx = sin dx + dx , sin l 0 l l l l l/2
2 bn = l
l
which is, integrating by parts, l/2 l/2 nπx nπx l x cos dx cos − − nπ l l 0 0 l l nπx nπx + (l − x) cos dx + cos l l/2 l l/2 8h nπ = sin . 2 2 n π 2
4h = 2 l
(B-1.81)
Thus, when h = 1 and l = π, the Fourier sine series on the interval 0 < x < π for the function f (x) is given by 1 1 8 (B-1.82) f (x) = 2 sin x − 2 sin 3x + 2 sin 5x − · · · . π 3 5 Thus, the Fourier sine series for f (x) = x on 0 < x < π2 is 1 1 4 sin x − 2 sin 3x + 2 sin 5x − · · · . x= π 3 5 Putting x =
π 2
(B-1.83)
in (B-1.83) gives the numerical series 1 1 1 π2 1 = 2 + 2 + 2 + 2 + ··· . 8 1 3 5 7
(B-1.84)
The Fourier sine series (B-1.83) for x can be integrated from x to π2 term-by-term so that 1 π2 1 1 4 cos x − 3 cos 3x + 3 cos 5x − · · · . (B-1.85) − x2 = 2 4 π 3 5 Substituting x = 0 into (B-1.85) gives the numerical series
B-1 Fourier Series and Its Basic Properties
1 1 1 π3 1 − 3 + 3 − 3 + ··· = . 3 1 3 5 7 32 Integrating (B-1.85) with respect to x from 0 to x leads to 4 1 x3 1 1 π2 = sin x − 4 sin 3x + 4 sin 5x − · · · . x− 2 4 3 π 3 5 Putting x =
π 2
737
(B-1.86)
(B-1.87)
in (B-1.87) yields the numerical series 1 1 1 π π3 1 π4 + + + + · · · = · = . 14 34 54 74 8 12 96
Integrating (B-1.87) again from x to
π 2
(B-1.88)
gives
2 4 1 π π − x2 − − x4 2 24 2 1 4 1 = cos x − 5 cos 3x + 5 cos 5x − · · · . π 3 5
π2 16
(B-1.89)
Substituting x = 0 in (B-1.89) gives the numerical series 1 1 1 1 5π 5 . − 5 + 5 − 5 + ··· = 5 1 3 5 7 1536 Integrating (B-1.89) again from 0 to x yields a new numerical series 4 π π2 3 1 5 π4 x− x − x+ x 64 48 384 120 4 1 1 = sin x − 6 sin 3x + 6 sin 5x − · · · . π 3 5 Putting x =
π 2
(B-1.90)
(B-1.91)
in (B-1.91) leads to another numerical series 1+
1 1 π6 1 . + + + · · · = 36 56 76 960
(B-1.92)
We consider applications of Fourier series to differential equations. As a simple application, we discuss the periodic solution of a non-homogeneous simple harmonic oscillator x ¨ + ω 2 x = f (t),
(B-1.93)
where the forcing function f (t) has a Fourier series expansion ∞ 1 (An cos nt + Bn sin nt), f (t) = A0 + 2 n=1
the coefficients A0 , An , and Bn are known.
(B-1.94)
738
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
We seek a uniformly convergent Fourier series solution of (B-1.93) in the form x(t) =
∞ 1 a0 + (an cos nt + bn sin nt). 2 n=1
(B-1.95)
Differentiating (B-1.95) term by term and assuming the series for x(t) ˙ and x ¨(t) converge uniformly, substituting in (B-1.93) gives ∞
∞ 2 a0 −n2 + ω 2 an cos nt + ω02 + −n + ω 2 bn sin nt 2 n=1 n=1 ∞ 1 (An cos nt + Bn sin nt). (B-1.96) = A0 + 2 n=1
Consequently, equating the coefficients, we obtain a0 =
A0 , ω2
an =
An , − n2
bn =
ω2
Bn . − n2
ω2
Thus, the solution of (B-1.93) is given by ∞
(An cos nt + Bn sin nt) A0 x(t) = + 2ω 2 n=1 ω 2 − n2
(B-1.97)
provided ω = n. However, if ω 2 = n2 for some integer n, the solution becomes unbounded. This phenomenon is known as resonance. If damping is included in equation (B-1.93), the solution will remain bounded. However, when ω 2 = n2 , we can write g(t) = f (t) − Aω cos ωt + Bω sin ωt, ¨ + ω 2 x = Aω cos ωt + and then we can solve separately x ¨ + ω 2 x = g(t) and x Bω sin ωt. The second equation gives rise to the non-periodic solutions (At + B) cos ωt and (A t + B ) sin ωt. The sum of the solutions of the two equations gives a solution of the original equation. Equation (B-1.93) can be solved by the use of integrating factor in the form τ t x(t) = exp(−iωt) exp(2iωτ ) eiωx f (x) dx dτ. (B-1.98) 0
0
We next apply the method of Fourier series to solve the damped harmonic oscillator governed by x ¨ + k x˙ + ω 2 x = f (t),
(B-1.99)
where k x˙ (k > 0) is the damping term. If f (t) is periodic with period 2π and has the same Fourier series expansion (B-1.94), then we can find a Fourier series solution of (B-1.95) by differentiating and substituting in (B-1.99) so that equation (B-1.99) takes the form
B-1 Fourier Series and Its Basic Properties ∞
−n an + nkbn + ω an cos nt + 2
n=1
+
2
ω 2 a0 2
739
∞ 2 −n bn − nkan + ω 2 bn sin nt n=1
∞ 1 (An cos nt + Bn sin nt). = A0 + 2 n=1
(B-1.100)
0 Equating the coefficients from both sides gives a0 = A ω 2 and 2 ω − n2 an + nkbn = An , −nkan + ω 2 − n2 bn = Bn .
Solving for an and bn gives An nk Bn ω 2 − n2 1 2 = an = 2 ω − n2 An − nkBn , 2 D ω − n 2 nk 2 −nk ω − n 2 ω − n2 An −nk Bn 1 2 = ω − n2 Bn + nkAn , bn = 2 2 D ω − n nk −nk ω 2 − n2
(B-1.101)
(B-1.102)
where 2 D = ω2 − n2 + n2 k 2 .
(B-1.103)
Thus, the coefficients a0 , an , and bn of the Fourier series solution (B-1.95) of the damped simple harmonic equation (B-1.99) are given by (B-1.101)–(B-1.102), and the solution represents the periodic response. There may be a transient response depending on the initial conditions, and the transient solution will eventually decay as t → ∞. Because of the presence of the damping term, there will be no resonance when n = ω. We close this section by adding an example of an application of Fourier series to a simple boundary value problem d2 u = λu = f (x), dx2 u(0) = 0 = u(l),
0 < x < l,
(B-1.104) (B-1.105)
where λ is a given parameter. For simplicity, we assume that f (x) has a Fourier sine series expansion ∞ nπx f (x) = , 0 < x < l, (B-1.106) bn sin l n=1
740
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
where bn are known. We seek the Fourier sine series solution of the boundary value problem (B-1.104)–(B-1.105) in the form ∞ nπx , 0 < x < l. (B-1.107) u(x) = Bn sin l n=1 We assume that this series may be differentiated twice to give ∞ nπx d2 u n2 π 2 = sin − B , 0 < x < l. n dx2 l2 l n=1 We substitute u(x) and u (x) into (B-1.104) to obtain ∞ ∞ nπx nπx n2 π 2 bn sin − 2 + λ Bn sin = , l l l n=1 n=1 Consequently, the coefficients Bn are given by −1 n2 π 2 Bn = λ − 2 bn , l
0 < x < l.
(B-1.108)
2 provided λ = ( nπ l ) . nπ 2 If λ = ( l ) for all or some values of positive integer n, then Bn cannot be determined, and hence no solution exists unless Bn ≡ 0. Thus, the Fourier series solution of the boundary value problem is ∞ l2 bn nπx sin , (B-1.109) u(x) = λl2 − n2 π 2 l n=1
where the zero denominator must be handled separately.
B-2 Generalized Functions (Distributions) The most widely known example of a generalized function is the Dirac delta function δ(x) which was first introduced by P.M.M. Dirac in 1920s as a mathematical device in the formulation of quantum mechanics. The Dirac delta function δ(x) is defined by ∞ δ(x) = 0 for all x = 0, and δ(x) dx = 1. (B-2.1) −∞
In the ordinary sense, δ(x) cannot be considered as a function because if a function vanishes everywhere except at a single point x = 0, then its integral over any interval must be zero, so that integrating it over an interval including the origin cannot be equal to one. However, from the physical point of view, the delta function is the natural mathematical quantity which can be used to describe many of the abstractions which arise in physical sciences. For example, the mass-density function
B-2 Generalized Functions (Distributions)
741
Fig. B.8 The sequence of delta functions, δn (x).
is zero everywhere except at x = 0, where it is infinite because a finite mass is concentrated in zero length, and it is so infinitely large here that the integral is non-zero even through the integrand is positive over an infinitesimally small region only. This makes sense physically, though it is mathematically absurd. So the delta function δ(x) can be used as the mass-density function ρ(x) describing the mass distribution per unit length of a rod at a point x. Similarly, the point charge, the impulsive force, the point dipole, and the frequency response of an undamped harmonic oscillator are all aptly represented by the delta function or other generalized functions. In general, the generalized functions play a major role in Fourier analysis and in the theory of partial differential equations. The function f (x) = 1 has no Fourier transform in the ordinary Fourier transform theory, but it has Fourier transform √ 2πδ(k) in the generalized function theory. Thus, the generalized functions remove difficulties which existed in the classical Fourier analysis. There is a beautiful analogy with the way in which the use of complex numbers helps in the solution of quadratic equations as within the realm of real numbers only certain quadratic equations have no solutions. Sometimes, the Dirac delta function δ(x) is defined by its fundamental property ∞ f (x)δ(x − a) dx = f (a), (B-2.2) −∞
where f (x) is any continuous function in any interval containing the point x = a. Although there are no ordinary functions which have the properties required of the delta function, we may approximate δ(x) in one dimension by a sequence of ordinary functions n exp −nx2 , n = 1, 2, 3, . . . . (B-2.3) δn (x) = π
742
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
Clearly, δn (x) → 0 as n → ∞ for any x = 0 and δn (0) → ∞ as n → ∞ as shown in Figure B.8. Also, for all n = 1, 2, 3, . . . , ∞ δn (x) dx = 1 −∞
and
lim
n→∞
∞ −∞
δn (x) dx =
∞
δ(x) dx = 1, −∞
as expected. Thus, the delta function can be considered as the limit of a sequence of ordinary Gaussian functions, and we write n exp −nx2 . (B-2.4) δ(x) = lim n→∞ π As mentioned in Chapter 1, the major problem of finding the solution of inhomogeneous partial differential equations deals with the construction of the Green’s function in each case. This problem becomes easier by the use of a generalized function together with the methods of Fourier and Laplace transforms. In spherical polar coordinates (r, θ, φ), the Laplacian is 1 d 2 2 dψ r , (B-2.5) ∇ ψ(r) = 2 r dr dr which is independent of θ and φ. If ψ(r) = 1r , then ∇2 ψ(r) = 0 for all r = 0. At r = 0, ∇2 ψ(r) does not exist. By the divergence theorem (or Gauss’ theorem) 1 1 2 1 ∇ div ∇ ∇ dV = dV = · n dS, (B-2.6) r r r V V S where V is the volume of a sphere of radius a and center at r = 0 so that d 1 eˆr 1 = eˆr = − 2, ∇ r r=a dr r r=a a
(B-2.7)
where eˆr is the direction of the outward normal n to the surface of the sphere. It follows from (B-2.6) that 1 2 1 dV = − ∇ eˆr · n ds = −4π. (B-2.8) r a2 S V Thus, the Laplacian ∇2 ( 1r ) is a function of r which has the following fundamental properties: (i) It vanishes for r = 0. (ii) It is not defined at r = 0. (iii) And its integral over any sphere with center at r = 0 is −4π.
B-2 Generalized Functions (Distributions)
All of these lead to the result −4π 2 1 dV = ∇ r 0 V
if V contains r = 0, if V does not contain r = 0.
This may be written in the compact form 2 1 ∇ = −4πδ(r), r where δ(r) is the Dirac delta function with the property 1 if V contains r = 0, δ(r) dV = 0 otherwise. V
743
(B-2.9)
(B-2.10)
(B-2.11)
This result is a special case of the general definition of the vector form of the delta function f (a) if V contains the point a, (B-2.12) f (r)δ(r − a) dr = 0 otherwise, V obtained by setting f (r) = 1 and a = 0 in (B-2.12). For the results (B-2.11) and (B-2.12) to be valid, it is certainly sufficient that the functions be continuous and infinitely differentiable everywhere. Result (B-2.12) reduces to (B-2.2) in one dimension. We would not go into great detail, but refer to the famous books of Lighthill (1958) and Jones (1982) for the introduction to the subject of generalized functions. A good function, g(x), is a function in C ∞ (R) that decays sufficiently rapidly so that g(x) and all of its derivatives decay to zero faster than |x|−N as |x| → ∞ for all N > 0. In other words, suppose that for each positive integer N and n, lim xN g (n) (x) = 0,
|x|→∞
(B-2.13)
then g(x) is called a good function. Usually, the class of good functions is represented by S. The good functions play an important role in Fourier analysis because the inversion, convolution, and differentiation theorems as well as many others take simple forms with no problem of convergence. The rapid decay and infinite differentiability properties of good functions lead to the fact that the Fourier transform of a good function is also a good function. Good functions also play an important role in the theory of generalized functions. A good function of bounded support is a special type of good function that also plays an important part in the theory of generalized functions. Good functions also have the following important properties. The sum (or difference) of two good functions is also a good function. The product and convolution of two good functions are good functions. The derivative of a good function is a good function; xn g(x) is a good
744
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
function for all non-negative integers n whenever g(x) is a good function. A good function belongs to Lp (a class of pth power Lebesgue integrable functions) for every p in 1 ≤ p ≤ ∞. The integral of a good function is not necessarily good. However, if φ(x) is a good function, then the function g defined for all x by x g(x) = φ(t) dt −∞
∞
is a good function if and only if −∞ φ(t) dt exists. Good functions are not only continuous, but are also uniformly continuous on R and absolutely continuous on R. However, a good function cannot necessarily be represented by a Taylor series expansion on every interval. As an example, consider a good function of bounded support exp[−(1 − x2 )−1 ] if |x| < 1, g(x) = 0 if |x| ≥ 1. The function g is infinitely differentiable at x = ±1, as it must be in order to be good. It does not have a Taylor series expansion in every interval because a Taylor expansion based on the various derivatives of g at any point x satisfying |x| > 1 would lead to zero value for all x. For example, exp(−x2 ), x exp(−x2 ), (1 + x2 )−1 exp(−x2 ), and sech2 x are good functions, while exp(−|x|) is not differentiable at x = 0, and the function (1 + x2 )−1 is not a good function as it decays too slowly as |x| → ∞. A sequence of good functions, {fn (x)}, is called regular if, for any good function g(x), ∞ fn (x)g(x) dx (B-2.14) lim n→∞
−∞
1 n φ(x)
exists. For example, fn (x) = is a regular sequence for any good function φ(x), since ∞ 1 ∞ lim fn (x)g(x) dx = lim φ(x)g(x) dx = 0. n→∞ −∞ n→∞ n −∞ Two regular sequences of good functions are equivalent if, for any good function g(x), the limit (B-2.14) exists and is the same for each sequence. A generalized function, f (x), is a regular sequence of good functions, and two generalized functions are equal if their defining sequences are equivalent. Generalized functions are, therefore, only defined in terms of their action on integrals of good functions if ∞ ∞ f, g = f (x)g(x) dx = lim fn (x)g(x) dx = lim fn , g (B-2.15) −∞
n→∞
−∞
n→∞
for any good function g(x), where the symbol f, g is used to denote the action of the generalized function f (x) on the good function g(x), or f, g represents the number that f associates with g. If f (x) is an ordinary function such that
B-2 Generalized Functions (Distributions)
745
(1 + x2 )−N f (x) is integrable on (−∞, ∞) for some N , then the generalized function f (x) equivalent to the ordinary function is defined as any sequence of good functions {fn (x)} such that, for any good function g(x), ∞ ∞ fn (x)g(x) dx = f (x)g(x) dx. (B-2.16) lim n→∞
−∞
−∞
For example, the generalized function equivalent to zero can be represented by either φ(x) of the sequences { φ(x) n } and { n2 }. The unit function, I(x), is defined by ∞ ∞ I(x)g(x) dx = g(x) dx (B-2.17) −∞
−∞
for any good function g(x). A very important and useful good function that defines x2 )}. Thus, the unit function is the generalized function the unit function is {exp(− 4n that is equivalent to the ordinary function f (x) = 1. The Heaviside function, H(x), is defined by ∞ ∞ H(x)g(x) dx = g(x) dx. (B-2.18) −∞
0
The generalized function H(x) is equivalent to the ordinary unit function 0 if x < 0, (B-2.19) H(x) = 1 if x > 0, and since generalized functions are defined through the action on integrals of good functions, the value of H(x) at x = 0 does not have significance here. The sign function, sgn(x), is defined by ∞ 0 ∞ sgn(x)g(x) dx = g(x) dx − g(x) dx (B-2.20) −∞
−∞
0
for any good function g(x). Thus, sgn(x) can be identified with the ordinary function sgn(x) =
−1
if x < 0,
+1
if x > 0.
In fact, sgn(x) = 2H(x) − I(x), which can be seen as follows: ∞ ∞ 2H(x) − I(x) g(x) dx sgn(x)g(x) dx = −∞ −∞ ∞ ∞ =2 H(x)g(x) dx − I(x)g(x) dx −∞ −∞ ∞ ∞ g(x) dx − g(x) dx =2 0
−∞
(B-2.21)
746
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
∞
=
g(x) dx −
0
0
g(x) dx. −∞
The following results are also true xδ(x) = 0, δ(x − a) = δ(a − x).
(B-2.22) (B-2.23)
Result (B-2.23) shows that δ(x) is an even function. Clearly, the result x 1 if x > 0, = H(x) δ(y) dy = 0 if x < 0 −∞ shows that
d H(x) = δ(x). dx The Fourier transform of the Dirac delta function is ∞ 1 1 F δ(x) = √ e−ikx δ(x) dx = √ . 2π 2π
(B-2.24)
(B-2.25)
−∞
Hence,
∞ 1 1 = eikx dk. δ(x) = F −1 √ 2π −∞ 2π
(B-2.26)
This is an integral representation of the delta function extensively used in quantum mechanics. Also, (B-2.26) can be rewritten as ∞ 1 eikx dx. (B-2.27) δ(k) = 2π −∞ The Dirac delta function, δ(x), is defined so that for any good function g(x), ∞ δ, g = δ(x)g(x) dx = g(0). (B-2.28) −∞
Derivatives of generalized functions are defined by the derivatives of any equivalent sequences of good functions. We can integrate by parts using any member of the sequences and assuming g(x) vanishes at infinity. We can obtain this definition as follows: ∞ f (x)g(x) dx f ,g = −∞ ∞ ∞ f (x)g (x) dx = − f, g . (B-2.29) = f (x)g(x) −∞ − −∞
The derivative of a generalized function f is the generalized function f defined by
B-2 Generalized Functions (Distributions)
f , g = − f, g
747
(B-2.30)
for any good function g. The differential calculus of generalized functions can easily be developed with locally integrable functions. To every locally integrable function f , there corresponds a generalized function (or distribution) defined by ∞ f, φ = f (x)φ(x) dx, (B-2.31) −∞
where φ is a test function on R → C with bounded support (φ is infinitely differentiable and such that its derivatives of all orders exist and are continuous). The derivative of a generalized function f is the generalized function f defined by (B-2.32) f , φ = − f, φ for all test functions φ. This definition follows from the fact that ∞ f ,φ = f (x)φ(x) dx −∞ ∞ ∞ = f (x)φ(x) −∞ − f (x)φ (x) dx, = − f, φ −∞
which was obtained from integration by parts and using the fact that φ vanishes at infinity. It is easy to check that H (x) = δ(x), for ∞ ∞ H (x)φ(x) dx = − H(x)φ (x) dx H ,φ = −∞ −∞ ∞ ∞ φ (x) dx = − φ(x) 0 = φ(0) = δ, φ. =− 0
Another result is δ ,φ = − It is easy to verify that
∞ −∞
δ(x)φ (x) dx = −φ (0).
f (x)δ(x) = f (0)δ(x).
We next define |x| = xsgn(x) and calculate its derivative as follows. We have d d dx d |x| = xsgn(x) = x sgn(x) + sgn(x) dx dx dx dx d =x 2H(x) − I(x) + sgn(x) dx = 2xδ(x) + sgn(x) = sgn(x), which is true since sgn(x) = 2H(x) − I(x) and xδ(x) = 0.
(B-2.33)
748
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
Similarly, we can show that d sgn(x) = 2H (x) = 2δ(x). dx
(B-2.34)
Theorem B-2.1. The Fourier transform of a good function is a good function. Proof. The Fourier transform of a good function f (x) exists and is defined by ∞ 1 F f (x) = F (k) = √ e−ikx f (x) dx. (B-2.35) 2π −∞ Differentiating F (k) n times and integrating N times by parts, we get ∞ N (n) (−1)N 1 −ikx d n F (k) ≤ √ (−ix) e f (x) dx (−ik)N 2π dxN −∞ ∞ N d n 1 1 √ ≤ x f (x) dx. |k|N 2π −∞ dxN Evidently, all derivatives tend to zero as fast as |k|−N as |k| → ∞ for any N > 0, and hence F (k) is a good function. Theorem B-2.2. If f (x) is a good function with the Fourier transform (B-2.35), then the inverse Fourier transform is given by ∞ 1 f (x) = √ eikx F (k) dk. (B-2.36) 2π −∞ Proof. For any > 0, we have 2 1 F e− x F (−x) = 2π
∞
−∞
e−ikx− x
2
∞
eixt f (t) dt dx.
−∞
Since f is a good function, the order of integration can be interchanged to obtain ∞ ∞ − x2 2 1 F (−x) = f (t) dt e−i(k−t)x− x dx, F e 2π −∞ −∞ which is, by a similar calculation as that used in Example 1.7.1 in Chapter 1, ∞ 1 (k − t)2 f (t) dt. = √ exp − 4 4π −∞ Using the fact that √ we can write
1 4π
(k − t)2 dt = 1, exp − 4 −∞ ∞
B-2 Generalized Functions (Distributions)
2 F e− x F (−x) − f (k) · 1 ∞ 1 (k − t)2 =√ dt. f (t) − f (k) exp − 4 4π −∞
749
(B-2.37)
Since f is a good function, we have f (t) − f (k) ≤ maxf (x). t−k x∈R It follows from (B-2.37) that − x2 F e F (−x) − f (k) ∞ 1 (t − k)2 ≤√ maxf (x) dt |t − k| exp − 4 4π x∈R −∞ ∞ 2 1 = √ maxf (x)4 |α|e−α dα → 0 4π x∈R −∞ as → 0, where α = Consequently,
t−k √ . 2
∞ 1 f (k) = F F (−x) = √ e−ikx F (−x) dx 2π −∞ ∞ 1 eikx F (x) dx = √ 2π −∞ ∞ ∞ 1 ikx = e dx e−iξx f (ξ) dξ. 2π −∞ −∞ Interchanging k with x, this reduces to the celebrated Fourier integral formula ∞ ∞ 1 f (x) = eikx e−ikξ f (ξ) dξ dk. (B-2.38) 2π −∞ −∞ Hence, the theorem is proved. Example B-2.1. The Fourier transform of a constant function c is √ F{c} = 2πcδ(k).
(B-2.39)
In the ordinary sense, c F{c} = √ 2π
∞
e−ikx dx
−∞
is not a well defined integral; it diverges. However, it can be treated as a generalized x2 function, namely taking c = cI(x) and considering {exp(− 4n )} as an equivalent sequence to the unit function I(x) gives
750
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
Fig. B.9 Graph of the function fa (x).
∞ c x2 x2 = √ dx, F c exp − exp −ikx − 4n 4n 2π −∞ which, by Example 1.7.1, is √ √ n exp −nk 2 , = c 2n exp −nk2 = 2πc π √ √ = 2πcδn (k) → 2πcδ(k) as n → ∞, since {δn (k)} = { nπ exp(−nk 2 )} is a sequence equivalent to the delta function defined by (B-2.4). Example B-2.2. Show that F e−ax H(x) = √
1 , 2π(ik + a)
(B-2.40)
a > 0.
We have, by definition, 1 F e−ax H(x) = √ 2π
∞
exp −x(ik + a) dx = √
0
1 . 2π(ik + a)
Example B-2.3. By considering the function (see Figure B.9) fa (x) = e−ax H(x) − eax H(−x),
a > 0,
(B-2.41)
find the Fourier transform of sgn(x). In Figure B.9, the vertical axis (y-axis) represents fa (x) and the horizontal axis represents the x-axis.
B-3 Basic Properties of the Fourier Transforms
751
We have, by definition, 0 ∞ 1 1 exp (a − ik)x dx + √ exp −(a + ik)x dx F fa (x) = − √ 2π −∞ 2π 0 2 (−ik) 1 1 1 . = √ − = · π a2 + k 2 2π a + ik a − ik In the limit as a → 0, fa (x) → sgn(x) and then 2 1 · . F sgn(x) = π ik
(B-2.42)
B-3 Basic Properties of the Fourier Transforms The Fourier transform of f (x), F{f (x)} = F (k), is defined by (1.7.1) and its inverse F −1 {F (k)} = f (x) is defined by (1.7.2). We state the following properties of the Fourier transform: (a) (Shifting)
(b) (Scaling)
(c) (Conjugate)
(d) (Translation)
(e) (Duality)
F f (x − a) = e−iak F f (x) ,
(B-3.1)
k 1 F f (ax) = F , |a| a
(B-3.2)
F f (−x) = F f (x) ,
(B-3.3)
F eiax f (x) = F (k − a),
(B-3.4)
F f (x) = F (k),
(f) (Composition)
∞
F F (x) = f (−k),
and
F (k)g(k)eikx dk =
−∞
where
∞
−∞
F (k) = F f (ξ)
and
f (ξ)G(ξ − x) dξ,
(B-3.5)
(B-3.6)
G(ξ) = F g(k) .
(g) (Riemann–Lebesgue Lemma) If F{f (x)} = F (k), then lim F (k) = 0.
|k|→∞
(B-3.7)
752
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
(h) (Poisson Summation Formula) √ ∞ 2π nπ . f (2πn) = F 2a a n=−∞ n=−∞ ∞
(B-3.8)
When a = π, ∞
∞ 1 f (2πn) = √ F (n). 2π n=−∞ n=−∞
(B-3.9)
When 2a = 1, formula (B-3.8) becomes ∞
f (n) =
∞ √ 2π F (n).
n=−∞
(B-3.10)
n=−∞
The convolution (f ∗ g)(x) of two integrable function f (x) and g(x) is defined by (1.7.13) and the convolution Theorem 1.7.1 states that F (f ∗ g)(x) = F (k)G(k), (B-3.11) or equivalently,
F −1 F (k)G(k) = (f ∗ g)(x).
(B-3.12)
Proof. We have 1 F (f ∗ g)(x) = √ 2π
∞ −∞
e−ikx (f ∗ g)(x) dx,
which is, by definition of the convolution, ∞ ∞ 1 e−ikx f (x − ξ)g(ξ) dξ dx = 2π −∞ −∞ ∞ ∞ 1 = e−ikξ g(ξ) dξ e−ik(x−ξ) f (x − ξ) dξ 2π −∞ −∞ ∞ ∞ 1 = e−ikξ g(ξ) dξ e−ikη f (η) dη, (x − ξ = η) 2π −∞ −∞ = F (k)G(k). This completes the proof. It is often convenient to delete the factor √12π in the definition of convolution (1.7.13) as this factor does not have any effect on the properties of the convolution. The convolution has the following algebraic properties: f ∗g = g∗f
(Commutative),
f ∗ (g ∗ h) = (f ∗ g) ∗ h
(Associative),
(B-3.13) (B-3.14)
B-3 Basic Properties of the Fourier Transforms
(αf + βg) ∗ h = α(f ∗ h) + β(g ∗ h) (Distributive), f ∗ δ = f = δ ∗ f (Identity),
753
(B-3.15) (B-3.16)
where α and β are any two constants, and δ(x) is the Dirac delta function. (f ∗ g)(x) = (f ∗ g)(x), x (f ∗ g)(x) = xf (x) ∗ g(x) + f (x) ∗ xg(x) , 1 h(ax + b + c), f (ax + b) ∗ g(ax + c) = |a|
(B-3.17) (B-3.18) (B-3.19)
where h(x) = (f ∗ g)(x). In particular, f (x + b) ∗ g(x + c) = h(x + b + c).
(B-3.20)
If a1 , a2 , b1 , b2 are any constants and f1 ∗ g1 , f1 ∗ g2 , f2 ∗ g1 , and f2 ∗ g2 exist, then (a1 f1 + a2 f2 ) ∗ (b1 g1 + b2 g2 ) = a1 b1 (f1 ∗ g1 ) + a1 b2 (f1 ∗ g2 ) + a2 b1 (f2 ∗ g1 ) + a2 b2 (f2 ∗ g2 ). (B-3.21) To prove (B-3.14), we have ∞ f ∗ (g ∗ h) (x) = f (x − ξ)(g ∗ h)(ξ) dξ −∞ ∞ ∞ = f (x − ξ) dξ g(ξ − t)h(t) dt −∞ −∞ ∞ ∞ f (x − ξ)g(ξ − t) dξ h(t) dt, (ξ − t = η) = −∞ −∞ ∞ ∞ f (x − t − η)g(η) dη h(t) dt, = −∞ −∞ ∞ = (f ∗ g)(x − t)h(t) dt = (f ∗ g) ∗ h (x). −∞
To prove (B-3.18), we apply the Fourier transform to the left hand side so that ∞ F x(f ∗ g)(x) = e−ikx x(f ∗ g)(x) dx −∞ ∞ ∞ −ikx = xe dx f (x − ξ)g(ξ) dξ −∞ −∞ ∞ ∞ −ikx = (x − ξ + ξ)e dx f (x − ξ)g(ξ) dξ −∞ −∞ ∞ ∞ (x − ξ)e−ikx dx f (x − ξ)g(ξ) dξ = −∞
−∞
754
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
∞ ∞ + e−ikx dx f (x − ξ)ξg(ξ) dξ −∞ −∞ ∞ ∞ e−ikx dx (x − ξ)f (x − ξ)g(ξ) dξ = −∞ −∞ ∞ ∞ + e−ikx dx f (x − ξ)ξg(ξ) dξ −∞ −∞ ∞ = e−ikx xf (x) ∗ g(x) dx −∞ ∞ + e−ikx f (x) ∗ xg(x) dx −∞ = F xf (x) ∗ g(x) + F f (x) ∗ xg(x) . We next apply F −1 to obtain result (B-3.18). To prove (B-3.19), we apply the Fourier transform to its left-hand side and use the convolution theorem so that F f (ax + b) ∗ g(ax + c) = F f (ax + b) F g(ax + c) 1 1 k k ikb ikc = F · G exp exp |a| a a |a| a a k 1 ik(b + c) H , since H(k) = F (k)G(k) = 2 · exp |a| a a 1 = F h(ax + b + c) . |a| We next apply F −1 to both sides to obtain (B-3.19). Theorem B-3.1 (General Parseval’s Relation). If F{f (x)} = F (k) and F{g(x)} = G(k), then ∞ ∞ f (x)g(x) dx = F (k)G(k) dk. (B-3.22) −∞
−∞
Proof. We proceed formally to obtain ∞ ∞ ∞ ∞ 1 ikx f (x)g(x) dx = dx e F (k) dk e−ilx G(l) dl 2π −∞ −∞ −∞ −∞ ∞ ∞ ∞ 1 = F (k) dk G(l) dl ei(k−l)x dx 2π −∞ −∞ −∞ ∞ ∞ F (k) dk G(l)δ(k − l) dl = −∞ −∞ ∞ = F (k)G(k) dk. −∞
Thus, the proof is complete.
B-3 Basic Properties of the Fourier Transforms
In particular, if f (x) = g(x), then ∞ f (x)f (x) dx =
∞
−∞
−∞
or equivalently,
∞ −∞
F (k)F (k) dk,
|f (x)| dx = 2
755
∞
−∞
|F (k)|2 dk.
(B-3.23)
This is well known as the Parseval relation. The function space L2 (R) of all complex-valued Lebesgue square integrable functions with the inner product (f, g) defined by ∞ (f, g) = f (x)g(x) dx (B-3.24) −∞
is a complex Hilbert space with the norm f 2 defined by ∞ 12 2 f 2 = (f, f ) = f (x) dx .
(B-3.25)
−∞
In terms of this norm, the Parseval relation (B-3.23) takes the form f 2 = F 2 = Ff 2 .
(B-3.26)
This means that the Fourier transform is a unitary transformation on the Schwartz space S(R) which consists of the set of all infinitely differentiable functions f so that f and all its derivatives are rapidly decreasing in the sense that sup |x|m f (n) (x) < ∞ for every m, n ≥ 0. x∈R
Physically, the quantity f 2 is a measure of energy and F 2 represents the power spectrum of f . The following examples of the Fourier transforms are useful in applied mathematics (see Debnath and Bhatta 2007). If χ[−a,a] (x) is the characteristic function of [−a, a] defined by 1 if |x| < a, χ[−a,a] (x) = H a − |x| = (B-3.27) 0 if |x| > a, then
2 sin ak . F χ[−a,a] (x) = Fa (k) = π k
If
|x| |x| H 1− , f (x) = 1 − a a
(B-3.28)
(B-3.29)
756
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
then
a F f (x) = √ 2π
ak 2
−2
2
sin
ak . 2
(B-3.30)
The following analytic properties of the convolution also hold: d (f ∗ g)(x) = f ∗ g (x) = f ∗ g (x), dx d2 (f ∗ g)(x) = f ∗ g (x) = f ∗ g (x), dx2 (f ∗ g)(m+n) (x) = f (m) ∗ g (n) (x), ∞ ∞ ∞ (f ∗ g)(x) dx = f (ξ) dξ g(η) dη, −∞
−∞ b
(f ∗ χ[a,b] )(x) =
−∞
(B-3.32) (B-3.33) (B-3.34)
x−a
f (x − ξ) dξ = a
(B-3.31)
f (η) dη,
(B-3.35)
x−b ∞
1 F f (x)g(x) = (F ∗ G)(k) = √ F (k − ξ)G(ξ) dξ, 2π −∞ f (x) = (f ∗ δ)(x) = F −1 F (k) .
(B-3.36) (B-3.37)
To prove (B-3.31), we have ∞ ∞ d (f ∗ g) (x) = f (x − ξ)g(ξ) dξ = f (x − ξ)g(ξ) dξ dx −∞ −∞ = f ∗ g (x). To prove (B-3.33), we first apply the Fourier transform to the left-hand side and then use the convolution Theorem 1.7.1 so that F (f ∗ g)m+n (x) = (ik)m+n F (f ∗ g)(x) = (ik)m+n F (k)G(k) = (ik)m F (k) (ik)n G(k) = F f (m) (x) F g (n) (x) = F f (m) ∗ g (n) (x) . The use of the inverse Fourier transform proves the result (B-3.33). To prove that the integral of the convolution satisfies (B-3.34), we have ∞ ∞ ∞ (f ∗ g)(x) dx = f (x − ξ)g(ξ) dξ dx −∞ −∞ −∞ ∞ ∞ g(ξ) f (x − ξ) dξ dξ, (x − ξ = η) = −∞ −∞ ∞ ∞ = g(ξ) dξ f (η) dη. −∞
−∞
We next prove (B-3.36) which is a dual result of (B-3.11) as follows. We have
B-3 Basic Properties of the Fourier Transforms
1 F f (x)g(x) = √ 2π
∞
757
e−ikx f (x)g(x) dx,
−∞
which is, by replacing g(x) by its inverse Fourier transform formula, ∞ ∞ 1 1 e−ikx f (x) dx √ eik x G k dk = √ 2π −∞ 2π −∞ ∞ ∞ 1 1 = √ G k dk · √ e−ix(k−k ) f (x) dx 2π −∞ 2π −∞ ∞ 1 = √ F k − k G k dk 2π −∞ = (F ∗ G)(k). Thus, the results (B-3.11) and (B-3.36) represent a duality since the Fourier transform of the convolution product of two functions is equal to the ordinary product of their Fourier transforms, and the Fourier transform of the ordinary product of two functions is equal to the convolution product of their Fourier transforms. If the diffusion kernel function Gt (x) is x2 1 exp − , (B-3.38) Gt (x) = √ 4κt 4πκt then (Gt ∗ Gs )(x) = Gt+s (x). To prove (B-3.39), we apply the Fourier transform without the factor 2 means that F{exp(−ax2 )} = πa exp(− k4a ). Consequently, F (Gt ∗ Gs )(x) = F Gt (x) F Gs (x) = exp −κk 2 t exp −κk 2 s = exp −k2 κ(t + s) = F Gt+s (x) .
(B-3.39) √1 . 2π
This
The inverse Fourier transform gives the result. 1 If g(x) = 2a H(a − |x|), then (f ∗ g)(x) is the average value of f (x) on [x − a, x + a]. Using the value of g(x) gives ∞ a 1 f (x − ξ)g(ξ) dξ = f (x − ξ) dξ (f ∗ g)(x) = 2a −a −∞ x+a 1 f (η) dη. = 2a x−a We close this section by adding an application of the convolution to the solution of the wave equation in Example 1.7.2. The solution of the Fourier transform U (k, t) = F{u(x, t)} of the transformed wave equation (1.7.17) is given by
758
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
U (k, t) = A cos ckt + B sin ckt, where
U (k, 0) = F (k)
and
dU dk
(B-3.40)
= G(k).
(B-3.41)
t=0
Consequently, the use of (B-3.41) gives U (k, t) = F (k) cos ckt +
G(k) sin ckt. ck
(B-3.42)
Using the results F δ(x − ct) + δ(x + ct) =
2 cos(ckt), π 2 sin ckt F χ[−ct,ct] (x) = F H ct − |x| = , π k applying the inverse Fourier transform to (B-3.42) gives the solution
1 sin ckt u(x, t) = F −1 F (k) cos ckt + F −1 G(k) . c k Application of the convolution Theorem 1.7.1 gives the solution π δ(x − ct) + δ(x + ct) u(x, t) = f (x) ∗ 2 1 π χ[−ct,ct] (x) + g(x) ∗ (B-3.43) c 2 ∞ 1 f (x − ξ) δ(ξ − ct) + δ(ξ + ct) dξ = 2 −∞ ∞ 1 + g(x − ξ)χ[−ct,ct] (ξ) dξ 2c −∞ ct 1 1 = f (x − ct) + f (x + ct) + g(x − ξ) dξ, (x − ξ = α) 2 2c −ct x+ct 1 1 g(α) dα. (B-3.44) = f (x − ct) + f (x + ct) + 2 2c x−ct This is identical with the d’Alembert solution of the wave equation.
B-4 Basic Properties of Laplace Transforms The Laplace transform of a continuous or piecewise continuous function f (t) for t > 0 is denoted by L{f (t)} = f (s), and it defined by (1.9.1) and the inverse
B-4 Basic Properties of Laplace Transforms
759
Laplace transform, L−1 {f (s)} = f (t), is defined by (1.9.2). In this section, some basic properties of the Laplace transform are presented. For more information, the reader is referred to Debnath and Bhatta (2007): (B-4.1) L eat f (t) = f (s − a) (shifting), b s 1 exp s f (scaling), (B-4.2) L f (at + b) = |a| a a dn L tn f (t) = (−1)n n f (s), n ≥ 1, (B-4.3)
∞ ds
f (t) = L f (s) ds, (B-4.4) t s
t
f (s) . (B-4.5) L f (τ ) dτ = s 0 If L{f (t)} = f (s), then L
−1
L−1
f (s) s f (s) s2
t
(B-4.6)
f (τ ) dτ,
= 0
t
t1
f (τ ) dτ
= 0
t
(t − τ )f (τ ) dτ.
=
(B-4.7)
0
0
In general, L−1
f (s) sn
t
t1
t2
= 0
0
=
0
t
0
···
tn−1
f (τ ) dτ dt1 · · · dtn−1
0
(t − τ )n−1 f (τ ) dτ. (n − 1)!
(B-4.8)
If f (t) is a periodic function with period T and L{f (t)} exists, then
−sT −1
L f (t) = 1 − e
T
e−st f (t) dt.
(B-4.9)
0
For example, if f (t) is a square wave function with period 2a defined by f (t) = H(t) − 2H(t − a) + 2H(t − 2a) − 2H(t − 3a) + · · · ,
(B-4.10)
then the Laplace transform of f (t) is f (s) =
1 as tanh . s 2
(B-4.11)
Clearly, the graph of f (t) shows that f (t) = 1 if 0 < t < a and f (t) = −1 if 0 < a < t < 2a. Thus, result (B-4.9) can be used to find
760
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
−1 f (s) = 1 − e−2as
2a
e−st f (t) dt 0 a 2a −2as −1 −st −st = 1−e e dt − e dt 0
a
1 · 1 − 2e−sa + e−2as = 1−e s 1 (1 − e−as )2 1 (1 − e−sa ) 1 as = . = = tanh s (1 − e−2as ) s (1 + e−sa ) s 2
−2as −1
Theorem B-4.1 (Convolution Theorem). If f (t) ∗ g(t) is the Laplace Convolution of f (t) and g(t) defined by t f (t) ∗ g(t) = (f ∗ g)(t) = f (t − τ )g(τ ) dτ, (B-4.12) 0
and if L{f (t)} = f (s) and L{g(t)} = g(s), then L f (t) ∗ g(t) = L f (t) L g(t) = f (s)g(s), or equivalently,
(B-4.13)
L−1 f (s)g(s) = f (t) ∗ g(t).
(B-4.14)
To prove the convolution theorem, we have t ∞ −st e f (t − τ )g(τ ) dτ dt, L f (t) ∗ g(t) = 0 0 ∞ t −s(t−τ ) = e f (t − τ ) · e−sτ g(τ ) dτ dt, 0
0
which is, reversing the order of integration, ∞ ∞ e−s(t−τ ) f (t − τ ) dt e−sτ g(τ ) dτ, = 0 ∞ τ ∞ = e−sx f (x) dx e−sτ g(τ ) dτ, (t − τ = x) 0 0 = L f (t) L g(t) = f (s)g(s). In case of the Laplace transform, we proved the convolution theorem (B-4.12) and now we prove the dual result 1 ¯ f (s) ∗ g¯(s), L f (t)g(t) = 2πi where 1 2πi c1 , c2 > 0.
f (t) = √
c1 +i∞ c1 −i∞
est f¯(s) ds,
g(t) = √
1 2πi
(B-4.15)
c2 +i∞
est g¯(s) ds, c2 −i∞
B-4 Basic Properties of Laplace Transforms
761
We have, by definition and replacing g(t) by its inverse Laplace transform, ∞ e−st f (t)g(t) dt L f (t)g(t) = 0 c2 +i∞ ∞ 1 = e−st f (t) dt · ezt g¯(z) dz 2πi 0 c2 −i∞ c2 +i∞ ∞ 1 = g¯(z) dz e−t(s−z) f (t) dt 2πi c2 −i∞ 0 c2 +i∞ 1 f¯(s − z)¯ g (z) dz, Re(s − z) ≥ c1 = 2πi c2 −i∞ 1 ¯ f (s) ∗ g¯(s), Re s ≥ c1 + c2 . = 2πi Like the above duality property for the Fourier transform, results (B-4.12) and (B-4.16) represent the duality property for the Laplace transform. The following properties of the Laplace convolution (B-4.12) also hold: L (f ∗ g) (t) = L f ∗ g (t) + f (0)g(s) = sf (s)g(s), (B-4.16) where the prime denotes the derivative with respect to t. Similarly, L (f ∗ g) (t) = L f ∗ g (t) + g(0)f (s) = sf (s)g(s).
(B-4.17)
Results similar to (B-4.16)–(B-4.17) can be proved for the second and the higher derivatives of (f ∗ g)(t). The Duhamel formulas follow from (B-4.16) and (B-4.17) as t L−1 sf (s)g(s) = f (0)g(t) + g(t − τ )f (τ ) dτ, (B-4.18) 0
or equivalently, −1
L
t
sf (s)g(s) = g(0)f (t) +
f (t − τ )g (τ ) dτ.
(B-4.19)
0
If fp (t) = tp−1 e−t , t > 0, then direct differentiation gives fp (t) = (p − 1)fp−1 (t) − fp (t).
(B-4.20)
It also follows that fp (t) ∗ fq (t) exists for all p, q > 0 and satisfies the following identities (fp ∗ fq )(t) = B(p, q)fp+q (t) (fp ∗ fq ) (t) = B(p, q) (p + q − 1)fp+q−1 (t) − fp+q (t) (fp ∗ fq ) (t) = fp ∗ fq (t) = (p − 1)fp−1 (t) − fp (t) ∗ fq (t),
(B-4.21) (B-4.22) (B-4.23)
= (p − 1)B(p − 1, q)fp+q−1 (t) − B(p, q)fp+q (t), (B-4.24) where B(p, q) are the Beta function of p and q.
762
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
To prove (B-4.21), we have t fp (t − τ )fq (τ ) dτ (fp ∗ fq )(t) = 0 t (t − τ )p−1 e−(t−τ ) τ q−1 e−τ dτ, = 0
= e−t tp+q−1
(t − τ = tu)
1
up−1 (1 − u)q−1 du, 0
= B(p, q)e−t tp+q−1 = B(p, q)fp+q (t). Then (t) = B(p, q) (p + q − 1)fp+q−1 (t) − fp+q (t) . (fp ∗ fq ) (t) = B(p, q)fp+q The Laplace convolution (B-4.12) also satisfies the properties similar to (B-3.13). In particular, (B-4.25) L (f1 ∗ f2 ∗ · · · ∗ fn )(t) = f 1 (s)f 2 (s) · · · f n (s). n ∗n (B-4.26) L f (t) = f (s) , where f ∗n is the nth convolution product defined by f ∗n (t) = (f ∗ f ∗ · · · ∗ f )(t).
(B-4.27)
In general, mathematical operations such as addition, f (x)+g(x), multiplication, f (x)g(x), and composition, f (g(x)), for two functions f (x) and g(x) form a new function or an ordinary output. On the other hand, the convolution f (x) ∗ g(x) represents the integral output of two functions f and g, and it plays a central role in the subjects, such as Fourier series, Fourier transforms, number theory, harmonic analysis, probability theory, and almost any integral transform. We have already stated the convolution Theorem 1.7.1 for the Fourier transform, and the convolution Theorem B-4.1 for the Laplace transform of two functions. Obviously, these theorems can be generalized to obtain a relation between the n-fold convolution of n functions and the product of the transforms of these functions. We next establish a nice connection between the Weierstrass transform and the two-sided Laplace transform of f (ξ) on −∞ < ξ < ∞ defined by ∞ e−sξ f (ξ) dξ (B-4.28) L f (ξ) = f (s) = −∞
which is the Fourier transform of f (ξ) for s = ik so that √ √ L f (ξ) = f (s) = 2πF f (ξ) (ik) = 2πF (ik). 2
(B-4.29)
ξ ) is given by Thus, the two-sided Laplace transform of f (ξ) = exp(− 4a
B-4 Basic Properties of Laplace Transforms
2 √ ξ2 ξ L exp − = 2πF e− 4a (ik = s) 4a √ = 4πa exp as2 .
763
(B-4.30)
In Example 1.7.6 in Chapter 1, the solution of the Cauchy problem for the diffusion equation is given by ∞ f (ξ)G(x − ξ) dξ = f (x) ∗ G(x), (B-4.31) u(x, t) = −∞
where the diffusion kernel G(x) is given by x2 1 . exp − G(x) = √ 4κt 4πκt
(B-4.32)
The above results (B-4.31)–(B-4.32) are used to introduce the Weierstrass transform of f (ξ) with positive parameter t in the form ∞ W f (ξ) (x) = F (x) = f (ξ)G(x − ξ) dξ, (B-4.33) −∞
where G(x), or, more precisely, Gt (x), is the kernel of the Weierstrass transform defined by (B-4.33). We next expand the kernel Gt (x − ξ) to express the Weierstrass transform (B-4.33) of f (ξ) in terms of the two-sided Laplace transform (B-4.28) so that (B-4.33) reduces to F (x) = W f (ξ) (x) 2 exp(− x ) ∞ x ξ2 exp − − ξ exp − f (ξ) dξ = √ 4κt 2κt 4κt 4πκt −∞ 2
exp(− x ) ξ2 x f (ξ) − , (B-4.34) = √ 4κt L exp − 4κt 2κt 4πκt where L is the two-sided Laplace transform (B-4.28). We use (B-4.32) with κt = a and (B-4.30) to derive the inverse Weierstrass transform so that 2 ∞ 1 ξ as2 −sξ e =√ e exp − dξ. (B-4.35) 4a 4πa −∞ Replacing s by the operator D = e−ξD f (x) =
and using the fact that
∞ (−ξ)k Dk f (x) k=0
equation (B-4.35) becomes
d dx
k!
= f (x − ξ),
(B-4.36)
764
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms 2
2 ∞ −ξD 1 ξ f (x) exp − dξ e 4a 4πa −∞ 2 ∞ 1 ξ dξ = √ f (x − ξ) exp − 4a 4πa −∞ ∞ (x − ξ)2 1 √ exp − = f (ξ) dξ 4a 4πa −∞ ∞ G(x − ξ)f (ξ) dξ = F (x). =
eaD f (x) = √
(B-4.37)
−∞
This gives the formula for the inverse Weierstrass transform f (x) = e−aD F (x), 2
x ∈ R.
(B-4.38)
The celebrated Riemann–Liouville fractional integral of order α of a function f (t) is usually defined by D
−α
f (t)
=0 Dt−α f (t)
1 = Γ (α)
t
(t − x)α−1 f (x) dx,
Re α > 0.
(B-4.39)
0
Clearly, D−α is a linear integral operator, and (B-4.39) can be expressed in terms of the convolution product D−α f (t) = f (t) ∗ g(t),
(B-4.40)
α−1
where g(t) = Γt (α) and g(s) = L{g(t)} = s−α . Using the Laplace convolution theorem to (B-4.40) gives L D−α f (t) = L f (t) ∗ g(t) = L f (t) L g(t) = s−α f (s),
(B-4.41)
or equivalently, D−α f (t) = L−1 s−α f (s) .
(B-4.42)
This can be used to evaluate the fractional integral of a given function f (t). For example, if f (t) = tβ , then
Γ (β + 1) Γ (β + 1) α+β t . (B-4.43) = D −α tβ = L−1 α+β+1 s Γ (α + β + 1) Another consequence of (B-4.39) for the fractional derivative, D−α f (t), is that it can be defined as the solution φ(t) of the integral equation D−α φ(t) = f (t) so that its Laplace transform gives
(B-4.44)
B-4 Basic Properties of Laplace Transforms
φ(s) = sα f (s). Consequently, the inverse Laplace transform yields φ(t) = D−α f (t) = L−1 sα f (s) t 1 t−α−1 ∗ f (t) = = (t − x)−α−1 f (x) dx. Γ (−α) Γ (−α) 0
765
(B-4.45)
(B-4.46) (B-4.47)
This is known as the Cauchy integral formula for the fractional derivative of f (t). In fact, (B-4.46) can often be used to obtain the fractional derivative of f (t). For example, if f (t) = tβ then
Γ (β + 1) (β−α) Γ (β + 1) t . (B-4.48) Dα tβ = L−1 = sβ−α+1 Γ (β − α + 1 In general, the Laplace convolution integral equation is of the form t f (t) = h(t) + λ g(t − τ )f (τ ) dτ,
(B-4.49)
0
where λ is a given constant parameter, f (t) is the unknown function, h(t) and g(t) are given functions. Applications of the Laplace transform to (B-4.49) combined with the convolution theorem yields f (s) =
h(s) . 1 − λg(s)
(B-4.50)
The inverse Laplace transform gives the formal solution of f (t) in the form
h(s) . (B-4.51) f (t) = L−1 1 − λg(s) For example, the Abel integral equation of the first kind in the form t (t − τ )α−1 f (τ ) dτ = g(t),
(B-4.52)
0
or equivalently, tα−1 ∗ f (t) = g(t),
(B-4.53)
can be solved by application of the Laplace transform so that f (s) = This leads to the solution
1 α 1 α−1 s g(s) = s s g(s) . Γ (α) Γ (α)
(B-4.54)
766
B Fourier Series, Generalized Functions, and Fourier and Laplace Transforms
f (t) =
1 d 1 Γ (α) Γ (1 − α) dt
t
(t − τ )−α g(τ ) dτ.
(B-4.55)
0
The Abel integral equation of the second kind is given by f (t) +
λ Γ (α)
t
(t − τ )α−1 f (τ ) dτ = g(t),
α > 0,
(B-4.56)
0
where λ is a real or complex parameter and g(t) is a given function. Application of the Laplace transform to (B-4.56) leads to the solution α sα−1 s f (s) = g(s) = s · g(s) , (B-4.57) · sα + λ sα + λ so that the inverse Laplace transform gives the solution of (B-4.56) as d t f (t) = Eα,1 −λτ α g(t − τ ) dτ, dt 0
(B-4.58)
where the Mittag-Leffler function, Eα,β (z), is given by Eα,β (z) =
∞
zn , Γ (nα + β) n=0
α, β > 0.
(B-4.59)
C Answers and Hints to Selected Exercises
1.15 Exercises 1. (a) A = 4, B = 5, C = 1, B 2 − 4AC = 9 > 0. Hyperbolic, B±
√
B 2 −4AC 2A
= 1,
1 . 4
Integrating gives y − x = c1 and y =
1 x 4
dy dx
=
+ c2 ,
uξη = 13 (uη − 83 ); α = ξ + η, β = ξ − η, uαα − uββ = 13 (uα − uβ − 83 ). (b) A = 2, B = −3, C = 1, B 2 − 4AC = 1 > 0. Hyperbolic, Integrating gives x + 2y = c1 , x + y = c2 ;
dy dx
= − 12 , −1.
uξη = η − ξ.
(c) Hyperbolic, ξ = y 2 − x2 = c1 , η = y − x = c2 ;
uξη + η1 uξ = 0.
(d) Hyperbolic for y < 0 and elliptic for y > 0. Parabolic for y = 0. √ √ For the hyperbolic case, ξ = x + 2 −y, η = x − 2 −y, y = − 14 (x − c)2 , and uξη =
1 (uη 2(ξ−η)
− uξ ).
√ For the elliptic case, α = x, β = 2 y; uαα + uββ = β1 uβ . (e) Elliptic for y > e−x , parabolic for y = e−x , and hyperbolic for y < e−x . (f) Hyperbolic for x < 0 and elliptic for x > 0. For x < 0; ξ, η = 32 y ± (−x)3/2 , (y − c) = ± 23 (−x)3/2 (cubic parabolas). uξη =
1 6(ξ−η) (uξ
− uη ).
1 β = −x3/2 ; uαα + uββ + ( 3β )uβ = 0, where α and β √ satisfy the Beltrami equations βx = − x αy , βy = √1x αx .
For x > 0, α =
3 2 y,
(g) Hyperbolic for |x| < 2|y|, elliptic for |x| > 2|y|, and parabolic for |x| = 2|y|. L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1,
768
C Answers and Hints to Selected Exercises
(h) Hyperbolic in the first and third quadrants, elliptic in the second and fourth quadrants. (i) Elliptic for |x| < a, parabolic for |x| = a, and hyperbolic for |x| > a. (j) ξ = y 2 , η = x2 ; 2ξη(uξξ + uηη ) + ηuξ + ξuη = 0. 2. (d) Elliptic for all x and y. In this case, the characteristics equations are
dy dx
=
±i sech2 x so that the characteristics are y ∓ i tanh x = constant so that α = y, β = tanh x. Thus, the canonical form of the given equation is uαα + uββ = 2β(1 − β 2 )−1 uβ . (f) Hyperbolic for all x and y. The characteristic equations are
dy dx
= ±sech2 x
so that the characteristics are y ∓ tanh x = const., ξ = y + tanh x, and η = y − tanh x. Thus, the canonical form of the given equation is uξη = (η − ξ) × (uξ − uη ){4 − (ξ − η)2 }−1 . (g)
dy dx
= cosec y; ξ = x + cos y, η = y; uηη = sin2 η cos ηuξ .
4. (a) A = u2 , B = 2ux uy , C = −u2 . Hence, B 2 − 4AC = 4(u2x u2y + u4 ) > 0 for all u(x, y). Hyperbolic for all u(x, y). (b) |∇u| = u2x + u2y . A = 1 − u2x , B = −2ux uy , C = 1 − u2y . Hence, B 2 − 4AC = −4 + 4(u2x + u2y ) = −4 + 4(Δu)2 > 0, = 0, or < 0 for |Δu| > 1, = 1, or, < 1. nπx 2 5. αn = ( nπ ), n = 0, 1, 2, 3, . . . ; un (x, t) = an exp(−αn kt) cos( ), and !∞ ) is the cosine Fourier series, where a0 = u(x, 0) = a0 + n=1 an cos( nπx 1 f (x) dx and an = 2 0 f (x) cos( nπx ) dx. 0
6. For λ = −α2 , u(x, y) = (A cos αx + B sin αx)(C cosh αy + D sinh αy), C A = 0, and α = nπ a , n = 1, 2, 3, . . . ; D = − tanh αb, and u(x, y) = a !∞ nπ(b−y) nπx , where an = (sinh nπb )−1 ( a2 ) 0 f (x) × n=1 an sin( a ) sinh a a
sin( nπx a ) dx. 7. Hint: We assume u(x, t) = X(x)T (t) = 0 and substitute in the given equation
to obtain XT (t) + c2 T X (x) = 0, or equivalently, − c12 TT =
X X
= λ4 .
1.15 Exercises
769
We have X = λ4 X, T = −λ4 c2 T , where λ4 is a separation constant. Thus, X(x) = A cosh λx + B cos λx + C sinh λx + D sin λx, T (t) = E cos(λ2 ct) + F sin(λ2 ct),
B = −A,
D = −C,
X(x) = A(cosh λx − cos λx) + C(sinh λx − sin λx). From conditions X( ) = 0 = X ( ), we obtain cos λ − cos λ sinh λ + sin λ
sinh α − sin α =0 cosh α − cos α
=⇒
cos λ cosh λ = 1.
8. We seek a nontrivial separable solution u(x, t) = X(x)T (t) so that sin α = 0, !∞ nπ 2 α = ( nπ n=1 [an cos{( ) ct} + ), n = 1, 2, 3, 4, . . .; sinh α = 0. u(x, t) = )2 ct}] sin( nπx ). bn sin{( nπ 9. an = 0 for all even n, an = ( n24π2 ), n = 1, 5, 9, . . . , and an = −( n24π2 ), n = 3, 7, 11, . . . . ! n2 π 2 κt Thus, u(x, t) = ∞ ) sin( nπx n=1 an exp(− 2 ). 11. Application of the Fourier transform, F{u(x, t)} = U (k, t), to the problem gives
d2 U dt2
+ k 4 U = 0, U (k, 0) = F (k), and Ut (k, 0) = 0. Thus, the solu-
tion of the transformed system is U (k, t) = F (k) cos(k 2 t). Consequently, the inverse Fourier transform together with the Fourier convolution yields the so∞ lution u(x, t) = F −1 {F (k) cos(k 2 t)} = √12π −∞ f (x − ξ)h(ξ, t) dξ, where F −1 {cos(k 2 t)} =
√1 2t
2
cos( x4t − π4 ) = h(x, t).
12. Apply the joint Fourier and Laplace transforms (1.9.12) to obtain U (k, s) = (s+α)F (k)+G(k) . (s2 +sα+c2 k2 )
Application of the joint inverse transforms combined with the
convolution of the Fourier transform gives the solution for u(x, t). ∞ 13. u(x, t) = √12π −∞ {A(k) exp[i(kx + ωt)] + B(k) exp[i(kx − ωt)]} dk, where √ 1 G(k)], and B(k) = 12 [F (k) − ω = c2 k 2 + a2 , and A(k) = 12 [F (k) + iω 1 iω G(k)].
770
C Answers and Hints to Selected Exercises
15. Hint: F −1 cos (atk2 ) = sin 17.
1 x2 √ [cos( 4at ) 2 at
2
x ± sin( 4at )].
P ∞ sin ωt φ(x, z, t) = − exp ikx + |k|z dk, ω 2 = g|k|, 2π −∞ ω 2 √ g P ∞ Pt gt η(x, t) = cos ωt exp(ikx) dk ≈ √ · 3/2 cos 2π −∞ 4x 2 2π x for gt2 4x. 19. iP exp(εt) ∞ (U k − iε) exp(|k|z + ikx) dk, 2πρ (U k − iε)2 − g|k| −∞ P exp(εt) ∞ |k| exp(ikx) dk η(x, t) = . 2 2πρ −∞ (U k − iε) − g|k|
φ(x, z, t) =
23. Using the Fourier transform, U (k, y) = F{u(x, y)}, gives the solution of the transformed system in the form U (k, y) = F (k) cos(k 2 y). The inverse Fourier transform together with the Fourier convolution gives the solution u(x, y) = ∞ ξ2 √1 f (x − ξ) cos( 4y − π4 ) dξ. 4πy −∞ 24.
u(x, y, t) =
1 2π
∞
−∞
∞
1 F (k, ) cos c k 2 + 2 2 t −∞ × exp i(kx + y) dk d .
¯ (k, s) = 25. Application of the joint Laplace and Fourier transform gives U ¯ s). The use of the inverse Laplace transform combined with (s + κk 2 )−1 Q(k, the convolution theorem yields U (k, t) = exp −κk 2 t ∗ Q(k, t) t exp −κk2 (t − τ ) Q(k, τ ) dτ. = 0
1.15 Exercises
771
Application of the inverse Fourier transform and its convolution theorem gives
t
u(x, t) =
F −1 P (k, t − τ )Q(k, τ ) dτ
0
t
p(x, t − τ ) ∗ q(x, τ ) dτ t ∞ 1 p(x − ξ, t − τ )q(ξ, τ ) dξ dτ = √ 2π 0 −∞ t ∞ 1 (x − ξ)2 − 12 dξ, = √ (t − τ ) dτ q(ξ, τ ) exp − 4κ(t − τ ) 4πκ 0 −∞
=
0
2
1 x exp(− 4κτ ). where p(x, τ ) = F −1 {exp(−κk2 τ )} = √2κτ ∞ 1 ikt 27. u(x, t) = 2π {F (k) cos(xα)+ G(k) dk, where α2 = α2 (k) = α sin(xα)}e −∞ 1 2 c2 (k
− iak − b).
28. Hint: Seek a solution of the form ψ(x, y, t) = φn (x, t) sin(nπy) with ψ0 (x, y) = ψ0n (x) sin(nπy), so that φn (x, t) satisfies the equation
2 ∂ ∂ φn [ ∂t ∂x2
− α 2 φn ] +
2 2 2 n β ∂φ ∂x = 0, where α = (nπ) +κ . Apply the Fourier transform of φn (x, t) with ∞ respect to x, and use Ψn (k, 0) = F{ψ0n (x)}. φn (x, t) = √12π −∞ Ψn (k, 0) ×
exp[i{kx − ω(k)t}] dk, where ω(k) = −βk(k 2 + α2 )−1 . 29. Divide the first equation by L and the second equation by C to obtain the third 1 G 1 and the fourth equations It + R L I = − L Vx , and Vt + C V = − C Ix . Differentiate
the third equation with respect to time t and replace Vt and Vx on the righthand side to derive a second order equation telegraph equation for I in the form Itt − c2 Ixx + (p + q)It + pqI = 0, where c2 = (LC)−1 , p =
R L,
and q =
G C.
Similar calculation can be used to obtain the same equation for V . Thus, u = I or V satisfies the above telegraph equation with coefficients c2 , p, and q, or with coefficients c2 , a, and b. 30. (a) V (x, t) = V0 erfc( 2√xκt ), κ = a−1 = (RC)−1 . I(x, t) =
Vo √ 1 R πκt
2
x ). exp(− 4κt
(b) V (x, t) = 12 e−x
√
b
erfc{ x2
a t
+
bt a}
+ 12 e−x
√
b
erfc{ x2
a t
−
bt a }.
×
772
C Answers and Hints to Selected Exercises
x x 31. V (x, t) = V0 exp(− kx c )f (t − c )H(t − c ).
32. u(x, t) = x cos ωt. 33. u(x, t) =
k [1 (πc)2
− cos( πct )] sin( πx ). a a 1
ω 2 ) (1 + i)z], u(z, t) = U erfc( 2√zνt ). 34. u(z, t)U exp[iωt − ( 2ν
35. u(z, t) = U t[(1 + 2ζ 2 )erfc(ζ) −
2ζ √ π
exp(−ζ 2 )], ζ =
z √ . 2 νt
36. q(z, t) =
1 a iωt −λ1 z erfc ζ − it(2Ω + ω) 2 e e 2 1 + eλ1 z erfc ζ + it(2Ω + ω) 2 1 b + e−iωt e−λ2 z erfc ζ − it(2Ω − ω) 2 2 1 + eλ2 z erfc ζ + it(2Ω + ω) 2 , 1
where λ1,2 = { i(2Ω±ω) } 2 . Thus, ν
q(z, t) ∼ a exp(iω − λ1 z) + b exp(−iωt − λ2 z),
δ1,2 =
ν |2Ω ± ω|
12 .
Q 43. (b) Hint: f˜(k) = ( πak )J1 (ak). ∞ 45. u(r, t) = 0 k f˜(k) cos(btk2 )J0 (kr) dk. 2 If f (r) = exp(− ar 2 ), then f˜(k) =
a2 2
2 2
exp(− a 4k ).
Using the self-reciprocity of the Hankel transform gives the solution 2 2 2 a k 1 2 cos btk (r) u(r, t) = a H0 exp − 2 4 = a2 Ω(t) exp −a2 Ω(t)r2 a2 cos 4btΩ(t)r2 + 4bt sin 4btΩ(t)r2 , where Ω(t) = (a4 + 16b2 t2 )−1 . 48. Hint: The solution of the dual integral equations ∞ kJ0 (kr)A(k) dk = u0 , 0 ≤ r ≤ a, 0 ∞ k 2 J0 (kr)A(k) dk = 0, a < r < ∞, 0
is given by A(k) = ( 2uπ0 ) sin(ak) k2 .
1.15 Exercises
773
49. Hint: See Debnath (1994, pp. 103–105). ∞ 1 ) 0 k −1 J1 (kr)J0 (kr) exp(−kz) dk. 50. u(r, z) = ( πa
√ 1 1 52. Hint: L−1 [(s2 + a2 )− 2 exp{−k(s2 + a2 ) 2 }] = H(t − k)J0 (a t2 − k 2 ). Application of the joint Laplace and Hankel transforms
∞
˜ u ¯(k, z, s) =
∞
rJ0 (kr) dr 0
e−st u(r, z, t) dt
0
to the given equation and the general boundary condition uz (r, 0, t) = f (r, t) = H(a − r)g(t) gives ˜¯ d2 u 1 ˜ ¯ = 0, − 2 s2 + k2 c2 u 2 dz c
˜ u ¯z (k, 0, s) =
a J1 (ak)¯ g (s). k
The bounded solution of the equation is z 2 ˜ 2 2 ¯ ˜ u ¯(k, z, s) = A(k, s) exp − s +c k c g¯(s)J1 (ak) ˜ ¯ s) = − ac √ with A(k, . k s2 + c2 k 2 √ J (ak) ˜¯(k, z, s) = −( ac √ 1 ¯(s) exp[− zc s2 + c2 k 2 ]. Thus, u k ) s2 +c2 k2 g The inverse Laplace transform gives g(s) ac z 2 −1 2 2 √ J1 (ak)L u ˜(k, z, t) = − exp − s +c k k c s2 + c2 k 2 t ac z z2 = − J1 (ak) g(t − τ )H τ − J0 ck τ 2 − 2 dτ. k c c 0 The inverse Hankel transform leads to the final solution
∞
t
u(r, z, t) = (−ac)
J0 (kr)J1 (ak) dk " # × J0 k c2 τ 2 − z 2 dτ. 0
0
z g(t − τ )H τ − c
774
C Answers and Hints to Selected Exercises
This, for g(t) = δ(t), becomes
z u(r, z, t) = (−ac)H t − c 53. u(r, z) = b
∞ 0
∞
J0 (kr)J1 (ak)J0 k c2 t2 − z 2 dk.
0
sinh kz k−1 ( cosh ka )J1 (bk)J0 (kr) dk.
54. Hint: H0 [(a2 − r2 )− 2 H(a − r)] = 1
sin(ak) , k
and
−1 √ √ √ s( s − a) exp −k s √ k = exp −ak − a2 t erfc √ − a t . 2 t
L−1
55. Hint: Use the joint Hankel and Laplace transform method. 57. Use the Hankel transform and derive 1 u(r, z, t) = ρ
∞
∞ r0 (t)
k exp(kz)J0 (kz) 0 0
× J0 (kα) dα cos ω(t − τ ) dτ ,
αp(α, τ )
0
where ω 2 = gk. 59. Hint: Use the joint Laplace and Fourier transform. W G(x, t) = 2πm where a2 =
EI m
∞
sin αt exp(ikx) dk, −∞ α
and ω 2 =
1 α = a2 k 4 + ω 2 2 ,
κ . m
61. Hint: Set G(x, y; ξ, η) =
∞ ∞
amn sin
m=1 n=1
mπx a
sin
nπy b
in the original equation to find amn from
mπ a
2 +
nπ b
2
mπξ ab nπη amn = sin sin . 4 a b
1.15 Exercises
775
62. Apply the joint Laplace and double Fourier transform. 63. Hint: Gtt − c2 Gxx + d2 G = δ(x)δ(t). The joint Laplace and Fourier transform gives 1 ˜ ¯ s) = √1 , G(k, 2 + α2 ) (s 2π
1 α = c2 k 2 + d2 2 .
The inverse transforms give − 12
∞ d2 d2 1 2 2 k + 2 sin ct k + 2 exp(ikx) dk G(x, t) = 2πc −∞ c c 1 d = c2 t2 − x2 H ct − |x| . J0 2c c 2 64. Hint: Eigenvalues are λn = ( nπ ) , n = 0, 1, 2, 3, . . . . Eigenfunctions are nπx Xn (x) = An cos( nπx ) + Bn sin( ).
Thus, u(x, t) =
∞
Xn (x)Tn (t) =
n=0
∞
un (x, t)
n=0
∞ nπx nπx nπkt 1 an cos + bn sin exp − , = a0 + 2 n=1 ∞ nπx nπx a0 + bn sin , + f (x) = u(x, 0) = an cos 2 n=1 where nπξ dξ, n = 0, 1, 2, . . . , f (ξ) cos − 1 nπξ f (ξ) sin bn = dξ, n = 1, 2, 3, . . . . −
1 an =
65. (b) Initial data tend to zero as n → ∞, but the solution un (x, y) → ∞ as n → ∞.
776
C Answers and Hints to Selected Exercises
66. (a) Hint: λ = 0, u(x) = A+Bx with the boundary conditions leads to the trivial solution. If λ < 0, u(x) = A cosh kx + B sinh kx(λ = −k2 ). The first boundary condition gives A = 0, and the second boundary condition yields k = − tanh k. The graphical representation of y = −k and y = tanh k shows that there is no intersection of these curves for k > 0. Hence, there are no negative eigenvalues. For λ = k 2 > 0, u(x) = A cos kx + B sin kx. The boundary conditions give A = 0, and B = 0, k = − tan k. The graphical representation of y = −k and y = tanh k shows that there is an infinite number of values k = kn (n = 1, 2, 3, . . .) so that there is an infinite number of eigenvalues λ = kn2 ≈
1 (2n 4
+ 1)2 π 2 for large n. The corresponding eigenfunctions are
un (x) = sin kn x, where n = 1, 2, 3, . . . . (b) λn = kn2 = (nπ)2 , un (x) = An cos nπx, n = 0, 1, 2, . . . . (c) λn = kn2 = (2nπ)2 , un (x) = {sin 2πnx, cos 2πnx}, (n = 0, 1, 2, . . .) are eigenfunctions. To each eigenvalue, there correspond two eigenfunctions. Eigenvalues are degenerate. (d) For 1 − 4λ = 0, or > 0, only trivial solutions. For 1 − 4λ = −k2 < 0, we obtain eigenvalues λn = 14 (1 + functions un (x) = Bn e−x sin kn x, (kn = 67. Hint: Multiply the equation by
p(x) a2 (x)
nπ a ,
n = 1, 2, 3, . . .).
so that
p(x)u + p (x)u + q(x) + λρ(x) u = 0, where p (x) =
p(x)a1 (x) , a2 (x)
q(x) =
p(x)a0 (x) , a2 (x)
n2 π 2 a2 )
and ρ(x) =
p(x) . a2 (x)
69. Hint: λ = k 2 > 0. The general solution is y(x) = A + Bx + C cos kx + D sin kx. The first boundary conditions at x = 0 give A = C = 0.
and eigen-
1.15 Exercises
777
The remaining boundary conditions at x = a give Ba + D sin ka = 0,
k2 D sin ak = 0.
For nontrivial solutions, B = C, D = 0, k = Thus, λn =
n2 π 2 a2 ,
nπ a .
and yn (x) = Dn sin( nπx a ). 2
2
The critical buckling loads are Pn = EI( naπ2 ). The Euler load is the largest load which the beam can withstand before possible buckling: P1 =
π2 a2 (EI).
The corresponding fundamental bucking mode is y1 (x) = D1 sin( πx a ). 71. X + 2bX + k 2 X = 0, T˙ + k2 T = 0, where −k2 is a separation constant. m = −b ± iα, α2 = k 2 − b2 > 0. X(x) = exp(−bx)(A cos αx + B sin αx). Using boundary conditions, we obtain A = 0, sin αa = 0, B = 0, α = ( nπ a ), n = 1, 2, 3, . . . . Hence, T (t) = C exp(−k 2 t). Thus, the final solution is u(x, t) =
∞
an exp −kn2 t − bx sin αn x,
kn2 = b2 + αn2 .
n=1
72. (a) Eigenvalues are k2 = ( λa )2 and eigenfunctions are
λr , Jn (ka) = 0, R(r) = AJn a λct λct T (t) = B cos + C sin , Θ(θ) = D cos nθ + E sin nθ, a a ∞ ∞ λmn r λmn ct (amn cos nθ + bnm sin nθ)Jn cos , u(r, θ, t) = a a n=0 m=0 where λmn are positive roots of Jn (x) = 0. λmn r , where a n=0 m=0 a π λm0 r 1 r dr dθ and n ≥ 1, = f (r, θ)J0 πa2 J12 (λn0 ) 0 −π a
f (r, θ) = a0n
∞ ∞
(anm cos nθ + bnm sin nθ)Jn
778
C Answers and Hints to Selected Exercises
(anm , bnm ) =
2 2 πa2 Jn+1 (λmn )
π a
−π 0
f (r, θ)Jn
λmn r a
× (cos nθ, sin nθ)r dr dθ. The set of frequencies is { πc a λmn : n ≥ 0, m ≥ 1}. (b) Since f is independent of θ, so is u. The solution is ∞ λm ct λm r cos , λm = λm0 , u(r, t) = am J0 a a m=1 ∞ λm r f (r) = a2 − r 2 = u(r, 0) = am J 0 a m=1 so that am =
8a2 λ3m J1 (λm ) .
73. (a) Hint: Use u(r, z) = R(r)Z(z) so that r2 R + rR + k 2 r2 R = 0,
R(a) = 0;
Z = k 2 Z,
Z(0) = 0.
The eigenvalues of this Sturm–Liouville problem are kn2 = (λn /a)2 with the corresponding eigenfunctions Rn (r) = J0 ( λan r ), where λn are the positive roots of J0 (x) = 0. !∞ Thus, Zn (z) = sinh( zλan ). u(r, z) = n=1 an J0 ( rλan ) sinh( zλan ), and g(r) = !∞ hλn rλn hλn u(r, h) = n=1 an sinh( a )J0 ( a ), where bn = an sinh( a ) = a rg(r)J0 ( rλan ) dr. 0 a 2cosech( hλn ) (b) an = cosech( rλan ) 0 rJ0 ( rλan ) dr = λn J1 (λna ) . Hence, the solution u(r, z) follows. 74. (a) Hint: Multiply the left-hand side of the wave equation by 2ut , and the resulting expression follows from performing the indicated differentiations on the right-hand side of the differential equality. (b) 2ut c2 ∇2n u − utt = 2c2 (ut ux1 )x1 + (ut ux2 )x2 + · · · + (ut uxn )xn − c2 u2x1 + u2x2 + · · · + u2xn + u2t t .
1.15 Exercises
779
(c) Perform the indicated differentiation on the right-hand side of (b) so that (c) can be obtained. 75. Hint: (a) In the context of a stretched string, −d2 u can be interpreted as an additional spring force normal to the string. Multiply the Klein–Gordon equation by ut to obtain ut utt − c2 ut uxx + d2 ut u = 0, ∂ 1 1 ∂ 2 c2 ∂ 2 u + u − c2 (ut ux ) + d2 u2 = 0. 2 ∂t t 2 ∂x x ∂x 2 (c) Assume w = u − v and u, v or (ux , vx ) are given at x = a, b. wtt = c2 wxx , a ≤ x ≤ b, w(x, 0) = 0 = wt (x, 0); w(orwx ) = 0 at x = a, b. Using (b), E(t) = E0 = 0, and noting w(x, 0) = 0 implies wx (x, 0) = 0. b Therefore, E(t) = 12 a (wt2 + c2 wx2 + d2 w2 ) dx = 0. Since the integrand is positive, w ≡ 0. 76. (a) Hint: Seek a separable solution u(r, t) = R(r)T (t) so that T + c2 k 2 T = 0, r 2 R + 2rR + k 2 r 2 R = 0, 1 R(r) = √ A1 J 12 (kr) + B1 Y 12 (kr) , r 1 T (t) = C cos(ckt) + D sin(ckt) = (A sin kr + B cos kr). r We assume that R(r) is finite at r = 0, and hence, B = 0. The eigenvalues are k = kn = nπ a , n = 1, 2, . . . . !∞ u(r, t) = n=1 (an cos ckn t + bn sin ckn t) 1r sin( nπr a ). The initial conditions u(r, 0) = f (r) and ut (r, 0) = g(r) can be satisfied by the Fourier series expansion of rf (r) and rg(r) on (0, a). (b) an = 0, b2n = 0, b2n−1 =
8a3 cπ4 (2n−1)4 .
780
C Answers and Hints to Selected Exercises
77. We seek a separable solution u(r, z) = R(r)Z(z) so that r2 R +rR −k 2 r2 R = 0, and Z + k2 Z = 0 with Z(0) = Z(h) = 0. k2 =
n2 π 2 h2
and Z(z) =
). We assume R(r) is finite at r = 0, and hence R(r) = an I0 ( nπr ), an sin( nπz h h !∞ ) sin( nπz ). We have f (z) = and thus, the solution is u(r, z) = n=1 an I0 ( nπr h h !∞ nπa nπz nπa −1 2 h nπz n=1 an I0 ( h ) sin( a ) so that an = [I0 ( h )] h 0 f (z) sin( h ) dz. !∞ 78. (a) Hint: u(1, θ) = a20 + n=1 (an cos nθ + bn sin nθ) = 1 + cos 2θ. a0 = 2, a2 = 1, an = 0, n = 0, 2. bn = 0 for all n, u(r, θ) = 1 + r2 cos 2θ. 0 π π (b) a0 = π1 −π |2θ| dθ = π2 [− −π θdθ + 0 θdθ] = 2π. π 8 an = π1 −π |2θ| cos θ dθ = − πn 2 for odd n, and 0 for even n. π bn = π1 −π |2θ| sin nθ dθ = 0. ! −2 2n+1 Thus, u(r, θ) = π − π8 ∞ r cos(2n + 1)θ. n=0 (2n + 1) !∞ (c) Hint: n=1 (an cos nθ + bn sin nθ) = 2 cos 2θ. a2 = 1, an = 0, n = 2; bn = 0 for all n. Therefore, u(r, θ) = (d) u(r, θ) =
a0 2
a0 2
+ r 2 cos 2θ.
+ r(cos θ + sin θ).
79. (a) Hint: Seek a separable solution u(r, θ) = R(r)Θ(θ) = 0 so that r2 R (r) + rR (r) − λ2 R = 0, Θ (θ) + λ2 Θ(θ) = 0. Since u(r, θ) is periodic with period 2π, so is Θ(θ). Eigenvalues: λn = λ = n, n = 1, 2, . . . . The solution of the Cauchy–Euler equation for R(r) is
n = 0:
R0 (r) =
1 (A0 + B0 ln r), 2
Θn (θ) = Cn cos nθ + Dn sin nθ, Rn (r) = An r n + Bn r −n , ∞ 1 an r n + bn r −n cos nθ u(r, θ) = (a0 + b0 ln r) + 2 n=1 + cn r n + dn r −n sin nθ. n ≥ 1:
1.15 Exercises
781
The Fourier coefficients for r = a are 1 π f (θ) dθ, (a0 + b0 ln a) = π −π π 1 an an + bn a−n = f (θ) cos nθ dθ, π −π π 1 cn an + dn a−n = f (θ) sin nθ dθ. π −π The process is similar, when r = b gives the Fourier coefficients. (b) Use the same method described in Exercise 76(a) to find r = 1:
a0 = 1,
an + bn = 0,
cn + bn = 0, r = 2:
1 c1 = − , 3 u(r, θ) =
d1 =
c1 + d1 = 1,
n > 1.
b0 = −1,
2n cn + 2−n dn = 0,
n ≥ 1;
2a1 +
b1 = 1, 2
n ≥ 1;
2n an + 2−n bn = 0
a1 =
2 , 3
for n > 1,
2 b1 = − , 3
4 , 3
1 2 1 (1 − ln r) + r − r −1 cos θ − r − 4r −1 sin θ. 2 3 3
(c) Use the solution obtained in Exercise 76(a) with the given boundary conditions. r = 1:
b0 = 0,
an − bn = 0,
r = 2:
b0 = 0,
1 3 a1 − b1 = , 4 4
n ≥ 1;
cn − dn = 0,
n ≥ 1,
1 3 c1 − d1 = . 4 4
Thus, a1 = b1 = 1 and c1 = d1 = −1. Hence, the solution is u(r, θ) =
a0 2
+ (r + r −1 ) cos θ − (r + r−1 ) sin θ.
81. Define v(x, y) = 14 (x2 + y 2 ) so that ∇2 φ = 1 and assume w = u + v so that ∇2 w = 0, w = v on ∂D, and w(0, 0) = u(0, 0). Use the result min∂D v ≤ u(0, 0) ≤ max∂D v.
782
C Answers and Hints to Selected Exercises
Max v =
a2 4
and the minimum of v =
circle x2 + y 2 = r2 which touches ∂D: 2
Since (x, y) = ( r4 ,
2
r 4
x a
r2 , 4
+
where r is determined from the y b
= 1, (x > 0, y > 0).
), one has r2 = a2 b2 (a2 + b2 )−1 .
82. Hint: We take v = u1 − u2 where u1 and u2 are solutions of ∇2 u = 0 on D. ∇2 v = 0 for x ∈ D and v(x) = 0 on ∂D. By the min–max principle, v attains its maximum and minimum on ∂D so that 0 ≤ v ≤ 0 for x ∈ D, and hence, v = 0 on D. 91. Hint: At the leading order ∇2 u0 = 0, u0 (1, θ) = sin θ. Seek a separable solution u0 = f (r) sin θ so that f + 1r f − r12 f = 0, f (1) = 1. The solution is f (r) = Ar + Br−1 . For the bounded solution, B ≡ 0. Hence f (r) = r, and u0 = r sin θ. At O(ε2 ), u2 satisfies the equation ∇2 u2 = −r sin θ,
u2 (1, θ) = 0. For the separable solution u2 = g(r) sin θ, g(r) satis-
fies g + 1r g −
1 r2 g
r = −A ,
g(1) = 0. Using the variation of parameters, the
bounded solution is g(r) = 18 (r − r3 ). 92. Hint: u = 0 satisfies the given equation and the far field boundary condition, but not the boundary condition at r = 1. We need a boundary layer near r = 1 so that we can define r = 1 + εr with r = O(1) in the boundary layer for ε 1. At
the leading order ur r − u = 0, which gives solution u = A(θ)er + B(θ)e−r . This matches with the far field if A ≡ 0 and satisfies the condition at r = 0 when B = 1. Thus the inner solution and a composite solution are valid at the leading order for r ≥ a. Hence the solution follows. 93. Hint: Seek a separable solution u(x, t) = X(x)T (t) = 0, where X(x) and T (t) satisfy X + λX = 0, 0 < x < , X (0) = 0 = X ( ), and T (t) + λκT = 0,
2 t > 0. The eigenvalues are λn = ( nπ ) and the eigenfunctions are
), n = 0, 1, 2, . . . . Thus, the solution is Xn (x) = An cos( nπx κn2 π 2 nπx u(x, t) = a0 + exp − 2 t , an cos n=1 ∞
1.15 Exercises ∞
nπx 1 f (x) = u(x, 0) = a0 + an cos , where a0 = n=1 nπx 2 dx, n = 1, 2, 3, . . . . an = f (x) cos 0 (b) a0 = 12 , an =
2 [(−1)n n2 π 2
783
f (x) dx, 0
− 1].
96. The solution follows from (1.9.15), when q(ξ, τ ) = sin(kξ − ωτ ). In case (a), when c =
ω , k
−1 u(x, t) = k 2 c2 − ω 2 sin(kx − ωt) −1 2 sin(kt + kct) + (kc − ω) 2kc ω − k 2 c2 2 −1 sin(kt − kct). + (kc + ω) 2kc ω − k 2 c2
Thus, the solution consists of three sinusoidal waves that propagate with different amplitudes and with velocities ±c and the phase velocity (ω/k). In case (b), when c =
u(x, t) =
ω k,
the solution is given by
1 1 t sin(x − t) − sin(x + t) + cos(x − t). 4 4 2
Thus, the solution consists of two harmonic waves that propagate with velocities ±1 and another harmonic wave whose amplitude grows linearly with time t. 98. Use the joint Laplace and finite Hankel transform of order one defined by
a
V¯ (km , s) =
∞
rJ1 (rkm ) dr 0
e−st v(r, t) dt,
0
where V¯ (km , s) is the Laplace transform of V (km , t), and km are the roots of the equation J1 (akm ) = 0. The solution of the transformed system is
784
C Answers and Hints to Selected Exercises
a2 νkm Ω f¯(s)J1 (akm ) . V¯ (km , s) = 2 ) (s + νkm (See Myint-U and Debnath 2007, Chapter 12, p. 507.) The solution is
v(r, t) = −2νΩ
∞ km J1 (rkm ) m=1
J1 (akm )
t
2 f (t − τ ) exp −νkm τ dτ.
0
When f (t) = cos ωt, the solution is given by
v(r, t) = −2νΩ
∞ km J1 (rkm ) m=1
J1 (akm )
t
2 cos ω(t − τ ) exp −νkm τ dτ.
0
When ω = 0, the solution becomes
v(r, t) = rΩ − 2Ω
∞ 2 J1 (rkm ) exp(−νtkm ) . k J (ak ) m 2 m m=1
In the limit as t → ∞, the transients die out and the ultimate steady-state is attained as the rigid body rotation about the axis of the cylinder. 99. (b) The Cauchy data tend to zero as n → ∞. But, for t > 0, the solution un (x, t) → ∞ as n → ∞ for certain values of x and t. So, the problem is ill-posed. 100. (b) The initial data tend to zero as n → ∞. However, for y = 0, the solution un (x, y) → ∞ as n → ∞. Thus, the problem is not well-posed. 101. (b) For y = 0, the amplitude of the solution tends to infinity as n → ∞ due √ to the factor sinh(ny), even though the initial data uy (x, 0) = exp(− n) × sin nx → 0 as n → ∞. However, the solution un (x, y) and all of its derivatives tend to zero as n → ∞ uniformly throughout the half-strip in the (x, y) plane. 102. Without the last term in the first equation, the equation is known as the Euler equation. The second equation is the continuity equation for a certain vir-
2.9 Exercises
785
tual fluid. The square of the amplitude a2 plays the role of density ρ and aV (x) plays the role of pressure. The second term on the right-hand side of the first equation plays the role of dispersion. Thus, the transformation of the Schrödinger equation into two equations in fluid dynamics is usually referred to as the hydrodynamic analogy of quantum mechanics. 103. (b) uxx = −n cos nx cosh ny, uyy = n cos nx cosh ny, and hence, uxx + uyy = 0. The boundary data is changed by only a small amount as u(0, y) → 0 as n → ∞. Yet the solution u(x, y) is changed from zero by a large amount. Along the line x = 0, the solution is u(0, y) =
1 n
cosh ny → ∞ as
y → ∞. Thus, a small change of the boundary data produces a large change in the solution. So, the problem is ill-posed. 104. (a) We have ut = κn2 exp(κn2 t) sin nx and uxx = −n2 exp(κn2 t) sin nx, and hence, ut + κuxx = 0. (b) Thus, u(x, 0) =
1 n
sin nx → 0 as n → ∞, yet u(x, t) → ∞ as n → ∞ for
any positive t. Hence, the problem is ill-posed. 105. (b) It is easy to verify that u(x, t) =
1 n
sin nx exp(κn2 −δn4 )t is the solution of
the equation. It also satisfies the initial condition. The solution is well behaved for large n, and also it is bounded for all n for any finite t. However, the solution is unstable as n → 0. For small δ, the negative diffusion equation is obviously ill-posed.
2.9 Exercises t t t 1. (i) Hint: 0 = δ t12 (T − V ) dt = t12 (δT − δV ) dt = t12 (m˙r · δ r˙ + F · δr) dt. t (ii) 0 = δ t12 ( 12 mx˙ 2 − 12 mω 2 x2 ) dt, or x ¨ + ω 2 x = 0. 2. Hint: Apply the variational principle t2 1 (T − V ) dt = 0, where T = ρ y˙ 2 dx and δ 2 0 t1 1 EIy 2 dx. V = 2 0
786
C Answers and Hints to Selected Exercises
Thus, ρ¨ y + EIy (iv) = 0. 3. ηt + 6ηηx + ηxxx = 0. 10. (a) We have ∇4 u = 0 (biharmonic equation). (b) utt − α2 ∇2 u + β 2 u = 0 (Klein–Gordon equation). (c) φt + αφx + βφxxx = 0 (φ = ux , KdV equation). (d) utt + α2 uxxxx = 0 (elastic beam equation). (e)
d (pu ) dx
+ (r + λs)u = 0 (Sturm–Liouville equation).
11. ( 2m )∇2 ψ + (E − V )ψ = 0. 2
14. utt − c2 uxx = F (x, t), c2 =
T∗ ρ .
18. Hint: (a) Use the vector identity (u · ∇)u = ∇( 12 u · u) − u × (∇ × u). (b) Use the fact that u · (u × ω ) = 0 to get the desired result. (c) Use ∇ · u = 0 and add the following result: ( 12 u · u +
p ρ
[u( 12 u · u +
+ Ω)(∇ · u) = 0 to the equation in (b) to find p ρ
∂ 1 ( u ∂t 2
· u) + ∇ ·
+ Ω)] = 0, where (u · ∇)φ + (u · ∇)φ = ∇ · (φu)·
Adding a zero contribution
∂Ω ∂t
gives the energy equation.
The first term represents the rate of change of the total energy (kinetic and potential), and the second term describes the energy flow carried by the velocity combined with the contribution from the rate of working of the pressure forces. 19. Hint: Follow Example 2.4.2 to obtain
1 + q2 + r2 px + 1 + p2 + r2 qy + 1 + p2 + q 2 rz − 2(pqpy + qrqz + rprx ) = 0.
22. (a) uy − ux y −
uy 1+y 2
= 0, (b) y 2 =
1−A2 (y1 −y) A2 (y1 −y) .
24. (a) Hint: For a conservative system, T + V = C. Putting V = C − T in (2.4.25) gives the principle of least action.
2.9 Exercises
787
(b) The principle of least action asserts that time action is stationary for any conservative system. 27. (a) In one dimension, the Lagrangian is L = T − V = ∂L ∂x
= − ∂V and ∂x
∂L ∂ x˙
1 mx˙ 2 2
= mx. ˙ The Euler–Lagrange equation is
∂L ∂x
− V (x) so that −
d ∂L ( ) dt ∂ x˙
= 0,
or equivalently, m¨ x = − ∂V ∂x = F . This is Newton’s law of motion. (b) The Lagrangian L = T − V = Lagrange equations are
d ∂L ( ) dt ∂ x˙
1 ˙2 2 m(x ∂L ∂x
=
and
+ y˙ 2 ) − V (x2 + y 2 ). The Euler– d ∂L ( ) dt ∂ y˙
=
∂L . ∂y
In this case, these
equations give m¨ x = − ∂V y = − ∂V ∂x and m¨ ∂y . In this formulation, conserved quantities are not obvious. (c) In polar coordinates (r, θ), x = r cos θ, y = r sin θ, x˙ = r˙ cos θ − rθ˙ sin θ, y˙ = r˙ sin θ + rθ˙ sin θ. Thus, x2 + y 2 = r2 and x˙ 2 + y˙ 2 = x˙ 2 + r2 θ˙ 2 . Thus, the Lagrangian becomes L = equations are dpθ dt
=
d ∂L ( ) dt ∂ r˙
d ˙ (mr2 θ) dt
mr, ˙ pθ =
∂L ∂ θ˙
=
∂L , ∂r
or,
= 0 because
˙ Since = mr2 θ.
1 2 2 m(r˙ dpr dt
∂L ∂r
dpθ dt
+ r2 θ˙ 2 ) − V (r2 ). The Euler–Lagrange
= m¨ r = mrθ˙2 −
= − ∂V + mrθ˙2 , ∂r
∂L ∂ θ˙
∂V ∂r
,
d ∂L ( ) dt ∂ θ˙
=
∂L , ∂r
or,
∂L ∂ r˙
=
= 0, and pr =
= 0, one gets mr 2 θ˙ = constant which shows
that the angular momentum is conserved and the radial equation of motion is m¨ r = mrθ˙ 2 −
∂V ∂r
. The first term in the radial equation on the right-hand side
represents the centrifugal force, while the second term gives the dynamical radial force. 28. In the linearized limit, f (r) = ae−brn − a ≈ −abrn so that the nonlinear Toda rn = ab(rn+1 + lattice equation m¨ r = 2f (rn )−f (rn−1 )−f (rn+1 ) becomes m¨ rn−1 − 2rn ). Substituting the traveling wave solution into the linearized Toda lattice equation gives mω 2 cos θ = 2ab cos θ(1 − cos k) = 4ab sin2 k2 . Thus, the dispersion relation is ω 2 =
4ab m
sin2 k2 , where k corresponds to the discrete
wavenumber that is similar to the continuous wavenumber.
788
C Answers and Hints to Selected Exercises
3.6 Exercises 2. (a) xp − yq = x − y, (d) yp − xq = y 2 − x2 . 3. (a) u = f (y), (b) u = f (bx − ay), (c) u = f (ye−x ), 2
(d) u = f (y − tan−1 x), (e) u = f ( x (f)
x+u y
−y 2 x ),
= C1 , (u + y)2 − x2 = C2 , f ( x+u , (u + y)2 − x2 ) = 0, y
x2 + y 2 = C1 , y(u − y) = C2 , (g) We have
dx y2
dy = − xy =
du xu−2xy
d(u−y) x(u−y) ,
=
u = y + y −1 f (x2 + y 2 ),
(h) u + log x = f (xy), (i) f (x2 + u2 , y3 + u3 ) = 0. 5. (a) u = sin(x − 32 y), (b) u = exp(x2 − y 2 ), (c) u = xy + f ( xy ), u = xy + 2 − ( xy )3 , (d) u = sin(y − 12 x2 ). ⎧ ⎪ ⎨ 12 y 2 + exp[−(x2 − y 2 )] for x > y, (e) u = ⎪ ⎩ 1 x2 + exp[−(y 2 − x2 )] for x < y. 2 (f) Hint: y = 12 x2 + C1 , u = C12 x + C2 , u = x(y − 12 x2 )2 + f (y − 12 x2 ), u = x(y − 12 x2 )2 + exp(y − 12 x2 ). (g)
y x
= C1 and
u+1 y
= C2 , C2 = 1 + dy −u
(h) Hint: x + y = C1 ,
=
du , u2 +C12
1 , C12
u=y+
x2 y
− 1, y = 0.
u2 + C12 = C2 exp(−2y).
From the Cauchy data, it follows that 1 + C12 = C2 , and hence,
u=
(i)
dy dx
u+1 x
−
y x
= 1,
d y dx ( x )
1 1 + (x + y)2 e−2y − (x + y)2 2 .
=
1 x
which implies that x = C1 exp( xy ).
y = C2 , f ( u+1 x , x exp(− x )) = 0.
Initial data imply x = C1 and
x2 +1 x
= C2 . Hence C2 = C1 +
1 C1 .
exp( xy ). Thus, u = x2 exp(− xy ) + exp( xy ) − 1. √ 6. We find u2 − 2ut + 2x = 0, and hence, u = t ± t2 − 2x. u+1 x
= x exp(− xy ) +
1 x
x 7. u(x, y) = exp( x2 −y 2 ).
3.6 Exercises
8. (a) u = f ( xy , xz ), (b) Hint: u1 =
x−y xy
= C1 ,
d(x−y) x2 −y 2
=
dz z(x+y)
gives u =
789 x−z z
=
x−y C2 , u = f ( x−y xy , z ). (c) φ = (x + y + z) = C1 . ( dx z ) y−z
Hint:
( dy y ) z−x
=
( dz z ) x−y
=
dy dx dz x + y + z
=
0
d log(xyz) , 0
=
ψ = xyz = C2 ,
u = f (x + y + z, xyz) is the general solution. (d) Hint: x dx + y dy = 0, x2 + y 2 = C1 , z dz = −(x2 + y 2 )y dy = −C1 y dy, z 2 + (x2 + y 2 )y 2 = C2 , u = f (x2 + y 2 , z 2 + (x2 + y 2 )y 2 ). (e)
x−1 dx y 2 −z 2
y −1 dy z 2 −x2
=
=
z −1 dz y 2 −x2
=
d(log xyz) . 0
Thus, u = f (x2 + y 2 + z 2 , xyz) is a
general solution. x2 2
9. (a) Hint: y − 3
u = xy −
+ f (y −
x 3
2
x 2
(b) u = xy − 13 x3 + y − 11.
x+u y
x3 3
= C1 , u = xy −
+ C2 , φ(u − xy +
), u = xy − x2 2
x 3
12. (a) x =
τ 2
2u y
= 0.
)2 .
− (x − y)2 − y2 (x − y) = 0.
+ τ s + s, y = τ + 2s, u = τ + s =
τ2 2
(b) x =
x 2
x2 2 )
+ (x − y), y > 0.
2 y
2
+ (y −
2
−
+ 56 .
= C1 , u2 − (x − y)2 = C2 , u2 −
Thus, u =
3
x3 3 ,y
(2x−2y+y 2 ) . 2(y−1)
+ τ s + s2 , y = τ + 2s, u = τ + s, (y − s)2 = 2x − s2 , a set of
parabolas. (c) x = 12 (τ + s)2 , y = u = τ + s. 13. Hint: The initial curve is a characteristic, and hence, no solution exists. 2
xy 14. (a) u = exp( x+y ), (b) u = sin[( x
−y 2 +1 12 ) ], 2
gives xy = c1 and u − ln y = c2 = f (xy). Hence, 2x − √ 1 y 1 2 ln(3x) = f (3x2 ). Or, f (t) = 2 3t − ln( 3t). Thus, u = 2( xy 3 ) + 2 log( 3x ), (c)
dx x
(e)
dx d
dy −y
=
=
=
dy y
du −1
=
du x2 +y 2
gives
x y
= c1 , and
xdx+ydy x2 +y 2
=
du x2 +y 2
yields 2u =
x2 + y 2 + c2 . Or, 2u = x2 + y 2 + f ( xy ). Hence, f (x) = x2 − 1. Thus, 2u = x2 + y 2 + (f)
dx y2
=
x2 y2
dy xy
− 1.
=
du x
gives x2 − y2 = c1 and u − ln y = c2 = f (x2 − y2 ). Hence,
f (t) = t + 1. Thus, u = ln y + (x2 − y 2 ) + 1.
790
C Answers and Hints to Selected Exercises dy y
(g)
dx x
xy 2
+ c2 . u =
=
=
du xy
xy 2
gives
dy x
= c1 and du = ydx = (c1 x)dx, u =
+ f ( xy ). Hence, f (1) = 0. Thus, u =
xy 2
c1 x2 2
+ c2 =
+ f ( xy ), where f is
an arbitrary function such that f (1) = 0. 15.
xdx x(u2 −y 2 )
And
=
ydy xy 2
udu −xu2
=
xdx+ydy+udu , 0
=
and hence, x2 + y 2 + u2 = c1 .
= − du gives uy = c2 . Thus, x2 + y 2 + u2 = f (uy), and hence, u
dy y
3u2 = f (u2 ). Thus, 3uy = u2 + x2 + y 2 . τ2 2
16. (a) x(s, τ ) = τ , y(s, τ ) =
+ aτ s + s, u(s, τ ) = τ + as. τ = x, s = (1 +
ax)−1 (y − 12 x2 )a, and hence, u(x, y) = x + as = (1 + ax)−1 {x + a(y + 12 x2 )}, singular at x = − a1 . (b) y =
u2 2
17. (a) Hint:
+ f (u − x), 2y = u2 + (u − x)2 , u(0, y) =
d(x+y+u) 2(x+y+u)
=
d(y−u) −(y−u)
d(u−x) −(u−x) ,
=
√
y.
(x + y + u)(y − u)2 = c1 and
(x + y + u)(u − x)2 = c2 . (b) Hint:
dx x
=
dy −y .
Hence, xy = a.
dx xu(u2 +a)
=
du x4 .
So,
dx du
=
u(u2 +a) x3
giving
x4 = u4 + 2au2 + b and, thus, x4 − u4 − 2u2 xy = b. (c)
dx x+y
=
dy x−y
dy 0
=
(exact equation). u = f (x2 − 2xy − y 2 ).
(d) f (x2 − y 2 , u − 12 y 2 (x2 − y 2 )) = 0. (e) f (x2 + y 2 + z 2 , ax + by + cz) = 0. 18. Hint:
dx x
=
dy y
=
dz x z , and hence, z
= c,
y z
= d. x2 +y 2 = a2 and z = tan−1 ( xy )
give (c2 + d2 )z 2 = a2 and z = b tan−1 ( dc ). c = ( az ) cos θ, d = ( az ) sin θ, and z = b tan−1 (tan θ) = bθ. Thus, the curves are x b θ = az cos θ and y b θ = az sin θ. 19. Hint:
(dx−2dy) 9u
=
du . −3(x−2y)
F {x + y + u, (x − 2y)2 + 3u2 } = 0. Thus,
(x − 2y)2 + 3u2 = (x + y + u)2 . 20. F (x2 + y, yu) = 0, (x2 + y)4 = yu. 21. Hint: x − y + z = c1 ,
dz −(x+y+z)
=
(dx+dy+dz) , 8z
and hence, 8z 2 + (x +
y + z)2 = c2 . F {(x − y + z), 8z 2 + (x + y + z)2 } = 0. c21 + c2 = 2a2 , or (x − y + z)2 + (x + y + z)2 + 8z 2 = 2a2 .
3.6 Exercises
22. F (x2 + y 2 + z 2 , y 2 − 2yz − z 2 ) = 0. (a) y 2 − 2yz − z 2 = 0, two planes are y = (1 ±
√
791
2)z.
(b) x2 + 2yz + 2z 2 = 0, a quadric cone with vertex at the origin. (c) x2 − 2yz + 2y 2 = 0, a quadric cone with vertex at the origin. 23. Use the Hint of 17(c). dx dt
= x + y,
dy dt
= x − y,
d2 x dt2
= 2x.
)2 = 2x2 + c. When x = 0 = y, ( dx dt √ 2 u = ln x + x2 − 2xy + 2y.
dx dt
=
√ 2x.
24. (a) a = f (x + 32 y). ). (b) x = at + c1 , y = bt, u = c2 ect , c2 = f (c1 ), u(x, y) = f (x − ab y) exp( cy b x )(1 − y)c . (c) u = f ( 1−y
(d) x = 12 t2 + αst + s, y = t, and u = y + 12 α(αy + 1)−1 (2x − y2 ). √ 26. (a) Hint: (f )2 = 1 − (g )2 = λ2 ; f (x) = λ and g (y) = 1 − λ2 . √ √ f (x) = λx + c1 and g(y) = y 1 − λ2 + c2 . u(x, y) = λx + y 1 − λ2 + c. (b) Hint: (f )2 + (g )2 = f (x) + g(y) or (f )2 − f (x) = g(y) − (g )2 = λ. Hence, (f )2 = f (x) + λ and (g ) = g(y) − λ. Or,
√ df f +λ
= dx and
√dg g−λ
= dy.
y+c2 2 1 2 f (x) + λ = ( x+c 2 ) and g(y) − λ = ( 2 ) . y+c2 2 1 2 u(x, y) = ( x+c 2 ) +( 2 ) .
(c) Hint: (f )2 + x2 = −g (y) = λ2 . √ Or, f (x) = λ2 − x2 , and g(y) = −λ2 y + c2 . Putting x = λ sin θ, we obtain √ f (x) = 12 λ2 sin−1 ( λx ) + x2 λ2 − x2 + c1 , √ u(x, y) = 12 λ2 sin−1 ( λx ) + x2 λ2 − x2 − λ2 y + (c1 + c2 ). (d) Hint: x2 (f )2 = λ2 and 1 − y 2 (g )2 = λ2 . √ Or, f (x) = λ ln x + c1 and g(y) = 1 − λ2 ln y + c2 . 27. (a) Hint: v = ln u gives vx =
1 u
· u x , vy =
1 u
· uy .
792
C Answers and Hints to Selected Exercises
x2 ( uux )2 + y 2 (
uy 2 u )
= 1.
Or, x2 vx2 + y2 vy2 = 1 gives x2 (f )2 + y 2 (g )2 = 1. x2 {f (x)}2 = 1 − y 2 (g )2 = λ2 .
√ Or, f (x) = λ ln x + c1 and g(y) = 1 − λ2 (ln y) + c2 . √ Thus v(x, y) = λ ln x + 1 − λ2 (ln y) + ln c, (c1 + c2 = ln c). u(x, y) = cxλ y
√ 1−λ2
.
(b) Hint: v = u2 and v(x, y) = f (x) + g(y) may not work. Try u = u(s), s = λxy, so that ux = u (y) · (λy) and uy = u (s) · (λx). 2 Consequently, 2λ2 ( u1 du ds ) = 1.
Or,
1 du u ds
=
√1 1 . 2λ
s Hence, u(s) = c1 exp( λ√ ). 2
xy ). Thus, u(x, y) = c1 exp( √ 2
28. Hint: vx =
ux 1√ 2 u,
vy =
uy 1√ 2 u.
This gives x4 (f )2 + y 2 (g )2 = 1.
Or, x4 (f )2 = 1 − y 2 (g )2 = λ2 . Or, x4 (f )2 = λ2 and y 2 (g )2 = 1 − λ2 . √ Hence, f (x) = − λx + c1 and g(y) = 1 − λ2 ln y + c2 . √ Thus, u(x, y) = (− λx + 1 − λ2 ln y + c)2 . 29. Hint: vx = (f )2 x2
ux , u
vy =
=1−
2 uy vx . u x2
+
vy2 y2
= 1, and v = f (x) + g(y).
1 2 y 2 (g )
= λ2 . √ f (x) = λx, and g (y) = 1 − λ2 y.
Or,
√ Or, f (x) = λ2 x2 + c1 , and g(y) = 12 y 2 1 − λ2 + c2 . 2√ v(x, y) = λ2 x2 + y2 1 − λ2 + c = ln u. 2√ u(x, y) = c exp[ λ2 x2 + y2 1 − λ2 ], c1 + c2 = ln c. 2
λ
2
ex = u(x, 0) = ce 2 x , which gives c = 1 and λ = 2. 31. (b) v(x, t) = x + at, u(x, t) =
6x+3at2 +5at3 . 6(1+2t)
32. (a) We have vt − avx = 0, v(x, 0) = ex . So,
dt 1
=
dx −a
=
dv 0
gives dv = 0,
or, v = const. = c1 . Also, dx + a dt = 0 yields x + at = const. = c2 . Thus,
3.6 Exercises
793
v = f (x + at), and ex = v(x, 0) = f (x). Or, v = ex+at . The first equation becomes ut + uux = e−x v = eat . This gives
dt 1
=
Or, u −
eat a
= c1 . Thus, dx = udt = ( ea + c1 )dt.
Or, x =
eat a2
+ c 1 t + c2 =
dx u
=
du eat .
at
Or, x − ut + at eat −
at
e a2
eat a2
+ (u −
eat )t a
eat a2
+ c2 =
= c2 . Thus, u(x, t) −
at
e a
+ ut −
teat a
= f (x − ut + at eat −
Hence, u − a1 eat = c1 and u(x, t) − a1 eat = f (x − ut + at eat − Using the initial condition gives x −
1 a
= f (x −
Hence, u(x, t) = a1 eat + (x − ut + at eat − Thus, u(x, t) = (1 + t)−1 [x + ( a1 +
1 ), a2
1 at a2 e
+ c2 .
+
1 at a2 e ).
or f (x) = (x + 1 a2
eat a2 ).
− a1 ).
1 a2
− a1 ).
− a12 )eat + ( a12 − a1 )]. √ dx du = dy = −u x − u1 = C1 and y = − ln(C2 u). 33. (a) We have √ 2 leads to 2 u x √ √ At (x0 , 0), u = 1 so that C1 = 2 x0 − 1 and C2 = 1. Thus, u−1 = 2( x − √ √ √ x0 ) + 1 and u = e−y , y = ln u−1 = ln[2( x − x0 ) + 1]. (b) We have
dx ux2
=
dy e−y
=
du −u2
t a
so that
1 x
= ln u + C1 and ey =
1 u
+ C2 .
At (x0 , 0), u = 1, that is, C2 = 0 and u = e−y . Hence, C1 =
1 x0 .
Thus, ln u = ( x1 −
34. For any function u(ξ, τ ), we have
1 x0 )
∂nu ∂xn
=
and y = ∂ n u ∂u ∂ξn , ∂t
1 x0
− x1 .
∂u = μ ∂u ∂τ − c ∂ξ ,
utt = μ2 uτ τ − 2μcuξ,τ + c2 uξξ , etc. Thus, the equation assumes the form uτ + (αun )uξ = 0. We have du 0
or
du dτ
= 0 and
dξ dτ
dτ 1
=
dξ αun
=
= αun . Thus, u = const. = C1 and ξ = αun τ + C2 .
Then the general solution is f (u, ξ − αun τ ) = 0, or u = φ(ξ − αun τ ) which is a Riemann simple wave. Hence, u(ξ, 0) = φ(ξ) = u0 sin kξ. Thus, u(ξ, τ ) = u0 sin[kξ − ( uu0 )n (αkun0 τ )]. 35. (a) We have
dx x+y
=
dy x−y
=
du 0 .
So, u = const. = c1 . Hence,
dy dx
=
x−y x+y .
This is
an exact equation and hence, f (x, y) = c2 , or, x2 − 2xy − y 2 = c2 . Thus, the general solution is u = f (x2 − 2xy − y 2 ), f is an arbitrary function.
794
C Answers and Hints to Selected Exercises dx 1
(b) We have
dy −(ax+by)
=
=
du 0 .
Thus, u = const. = c1 . Hence,
dy dx
+ by =
−ax, giving y(b2 y − abx + a) = c2 . Hence, u = f (y(b2 y − abx + a)) is the general solution. dx x−1
(c) We have
dy y −1
=
=
du . Hence, x2 −y 2 x2 −y 2
= c1 .
du dx
= x(x2 −y 2 ) = c1 x
giving u = 12 c1 x2 + c2 . Thus, u − 12 x2 (x2 − y 2 ) = c2 and u − 12 x2 (x2 − y 2 ) = f (x2 − y 2 ). dx y
(e) We find that
=
dy −x
=
du . 0
This gives x2 − y 2 = C1 and u = C2 . So,
u = f (x2 + y 2 ). Hence, y = f (a2 + y 2 ) = f (t), a2 + y 2 = t. Therefore, √ f (t) = t − a2 . Thus, u = f (x2 + y 2 ) = x2 + y 2 − a2 . dx x
(f) We find that dy dx
=
x+y , x
=
and hence,
dy x+y y x
= udu, or,
du dx
= log x + c2 . So,
= y x
1 , xu
or
u2 2
− log x = c1 . Thus, 2
− log x = f ( u2 − log x). Thus,
x − log x = f (− log x). Putting t = − log x, f (t) = t − e−t . Thus, 2
u 2
y x
=
2
− x exp(− u2 ).
36. (a) We have,
dt 1
=
dx −x
=
du u
gives t + ln x = C1 and xu = C2 .
Or, g(xu, t + ln x) = 0, or, u = x1 h(t + ln x) is the general solution, where h is an arbitrary function. Hence, u(x, 0) = f (x) = x1 h(ln x), or, h(ln x) = xf (x), that is, h(x) = ex f (ex ). Thus, u(x, t) = et f (xet ).
4.6 Exercises 1. (a) 16u = (x + 4y)2 , (b) u = α(x + y). √ 2. (a) a log u = x + ay + b, (b) 2 u(1 + a2 ) = x + ay + b, 1
(d) u = ay + 14 (x + b)2 , (g) u = a log x + (4 − a2 ) 2 log y + b, 1
(h) u = a(1 + x2 ) 2 + 12 (ay)2 + b. 3. (a) log u = x + y − 1, (c) u =
xy 2(y−2)
which is singular at y = 2,
(e) u(x, y) = x + y, (f) 4u = (x + y + 2)2 . (i) u = exp(−act)f (x − ct). Use equation (4.2.19) x(τ, 0) = x0 (τ ) = τ , t(τ, 0) = 0, u(τ, 0) = f (τ ).
4.6 Exercises
795
4. Hint: At s = 0, p(t, x) = q(t, s) = 1, then, p(t, s) = q(t, s) = 1. x(t, s) = (t − 2)es + 2, y(t, s) = 1 + es , u(t, s) = (t − 1)es + 2, u(x, y) = x + y − 1. 8. Hint: x(0, s) = 0, y(0, s) = s, u(0, s) = −s. x(t, s) = − sinh t, y(t, s) = s cosh t, 1
u(t, s) = − 2s (1 + cosh t), u(x, y) = −y(x2 + 1) 2 . 10. Hint: Characteristics are x − 2ct = s where s is a parameter. The points on the characteristic lines travel with speed 2c, whereas points on the wave profile u = c(x − ct) move with speed c. The wave profile u is not constant on the characteristic lines. 11. (a) u = 12 xy + f ( xy ) where f is an arbitrary function such that f (1) = 0. (b) Characteristics are xy = const. (a family of hyperbolas). u = a log y + f (xy), u =
a 2
1
y log( 3x ) + 2( xy )2 . 3
(c) At t = 0, p = 0, q = 1s ; sp = tanh( st ), sq = sech( st ). x(s, t) = −2s2 tan( st ) sech( st ),
y(s, t) = 2s2 sech2 ( st ) − s2 .
u(s, t) = 2s sech( st ), (u2 − 2y)2 = 4(x2 + y 2 ). (d) u(x, y) = x − y, (e) u(x, t) = x2 tan t. 12. Hint: Write the characteristic strip equations
dx dt
= Fp , . . . . Since
1 dp p dt
=
1 dq q dt ,
p and q are proportional to each other. The initial data p0 (s) and q0 (s) are 1
the solutions of p0 (s) = 0 and (p20 + q02 ) 2 = tan θ so that p0 (s) = 0 and q0 (s) = tan θ. Since p0 (s) = 0, the constant of proportionality must be zero, and so p(t, s) = 0. We then find x(t, s) = s and z(t, s) = cos(θ − t). Then
dq dt
= −[q 2 + q cos(θ − t)] gives q(t, s) = tan(θ − t)
y(t, s) =
− sin(θ − t) + 2 sin θ, or equivalently, (y − 2 sin θ)2 + z 2 = 1, which represents a cylinder. 2
1
1 1−z 2 14. Hint: ( dz dξ ) = ± z ( 1+a2 ) , 1
) = ±d(x cos θ + y sin θ). d(1 − z 2 ) 2 = ±d( √x+ay a2 +1
796
C Answers and Hints to Selected Exercises
15. (a) Hint: dx dt
dp dq
p q
=
gives p = c1 q.
du + (u − a) dp dt + p dt = 0, or dx + (u − a) dp + p du = 0,
or d(x + (u − a)p) = 0 gives x + p(u − a) = const. Similarly, y + q(u − a) = const. These equations and F = 0 give the integral surfaces (x−ct) 1+t(x−ct) .
x2 + y 2 + (u − a)2 − 1 = 0, u(x, t) = (b) Hint: dp dt
dx dt
= 2pu2 ,
= − 2p u , and
(pu ˙ + up) ˙ =
dq dt
dx dt
= 2qu2 ,
du dt
= 2(1 − u2 ),
= − 2q u .
d dt (up)
Consequently,
dy dt
= −2pu2 ,
d dt (uq)
= −2qu2 .
d = 2pu2 = − dt (up) and
dy dt
d = − dt (uq), and hence,
(x − a)2 + (y − b)2 = u2 (p2 + q 2 ) = 1 − u2 , where a and b are constants. (c) (x − x0 )2 + (y − y0 )2 = (u − u0 )2 , where x0 , y0 , and u0 are constants. This is a three-parameter ⎧ family of solutions. Use equation (4.3.1). ⎪ ⎨ x2 if y = 0, 16. (a) u(x, y) = ⎪ ⎩ (2y 2 )−1 (1 + 2xy − √1 + 4xy) if y = 0.
(b) u(x, y) =
⎧ ⎪ ⎨x
if y = 0,
⎪ ⎩ (2y)−1 (√1 + 4xy − 1)
if y = 0.
5.5 Exercises 1. (a) u(x, t) = sin ξ, ξ = x − t sin ξ. (b) u(x, t) =
x 1−t ,
u → ∞ as t → 1.
2. Hint: (u)t + ( 12 u2 )x = 0. Multiplying it by nun−1 gives the first result. 4. ξ = x − t exp(− 1ξ ) for x > 0; u(x, t) = exp(− 1ξ ). 6. The characteristic diagram in the (x, t)-plane shows that the initial signal f (x) is focused along the characteristics to a region near x = 0 as t increases.
5.5 Exercises
7. u = t + A, x = ut −
t2 2
797
+ B, where A and B are constants.
8. Hint: Shock t = 2(x − 1) from (1, 0) to (2, 2) and shock x = u = 0 for x < 0, and, when x ≥ 0, u =
x t
√
2t from (2, 2). √ to the left of t = x and x = 2t,
u = 0 to the right of shocks, and u = 1 in a triangular region bounded by t = 0, t = x, and t = 2(x − 1). 9. u = −1 to the left of the line t + x = 0, and u = 1 to the right of the line x − t = 1. Between these lines, u = (2x − 1)/(1 + 2t). 10. Characteristics from 0 ≤ x ≤ t, t = 0 intersect at ( 12 , 12 ). Characteristics from x ≤ 0, x ≥ 1 also intersect in the region t ≥
1 2,
and hence, no single-valued
solution. Shock parallel to the t-axis from ( 12 , 12 ). We find u(x, t) = (2x−1)/(1−2t) in the triangular region bounded by x−t = 0, x + t = 1, t = 0. u(x, t) =
⎧ ⎪ ⎨ +1 to the left of shock, ⎪ ⎩ −1 to the right of shock.
11. When a < 0, shock along the line at = 2x, −a to the left of shock, 2a to the right. When a > 0, −a to the left of the line x = −at and 2a to the right of the line x = 2at. Between these lines, u = ( xt ). √ dx = c(u) = 15. Hint: Here c(u) = u2 , f (x) = x, dt 1
du . 0
Thus, du = 0 gives u(x, t) = const. on the characteristics. √ u(x, t) = f (ξ) = ξ. √ ξ = x − tF (ξ) = x − tc(f (ξ)) = x − tc( ξ) = x − tξ. Thus, ξ =
x . 1+t
1
x Hence, u(x, t) = ( 1+t )2 . 1
16. c(u) = u3 , f (x) = x 3 , we have
dt 1
=
dx u3
=
du 0 .
u(x, t) = const. on the characteristics. 1
Thus, u(x, t) = f (ξ) = ξ 3 . 1
ξ = x − tF (ξ) = x − tc(f (ξ)) = x − tc(ξ 3 ) = x − tξ. Hence, ξ =
x 1+t
1
x and u(x, t) = ( 1+t )3 .
798
C Answers and Hints to Selected Exercises
17. c(u) = u, f (x) = x;
dt 1
dx u
=
=
du 2t
gives
du dt
= u and
du dt
= 2t. Hence,
u = t2 + A on the characteristic curve x = x(t) joining (x, t) and (ξ, 0). dx dt
Also,
= [u(x, t) − t2 ] + t2 = [u(ξ, 0) − 0] + t2 = ξ + t2 with ξ = x(0).
The equation of the characteristic curve is x(t) = ξt + x(t) = ξ(1 + t) +
3
t 3
Or,
dt 1
dx dt
=
dx 3t
=
du , u du u
= 3t and
+ B with x(0) = ξ.
. u(x(0), 0) = u(ξ, 0) = f (ξ) = ξ.
Thus, A = ξ, u(x, t) = t2 + ξ = t2 + 18. Hint:
t3 3
3x−t3 3(1+t) .
u(x, 0) = f (x) = cos x.
= dt. Or, x = 32 t2 + A and u(x, t) = Bet .
B = u(ξ, 0) = cos ξ, with x(0) = ξ. Then A = ξ and x − 32 t2 = ξ. Consequently, u(x, t) = et cos ξ = et cos(x − 32 t2 ). 19. Hint: dx dt
dt 2
=
dx 1
=
du , 0
and hence u(x, t) = const. on the characteristic curve
= 12 , or equivalently, x =
t 2
+ A.
If x(0) = ξ, then x = ξ + 2t . u = (x, t) = const. = u(ξ, 0) =
Or equivalently, u(x, t) = 20. Hint: du dt
dt 1
=
dx 1
=
du x
⎧ ⎪ ⎨ sin ξ
if 0 ≤ ξ ≤ π,
⎪ ⎩0
otherwise.
⎧ ⎪ ⎨ sin(x − 2t ) ⎪ ⎩0
if 0 ≤ x − 12 t ≤ π, otherwise.
gives x = t + A, and hence, x = t + ξ, if x(0) = ξ.
= x(t) = t + ξ, and hence, u(x, t) = 12 t2 + ξt + B.
Or, B = u(x, t) − 12 t2 − ξt = u(ξ, 0) = sin ξ. If x > t, u(x, t) = 12 t2 + t(x − t) + sin(x − t). If (x, t) lies above the line x = t, characteristic passing through this point never reaches the x-axis, but it reaches the t-axis so that the boundary condition u(0, t) = t can be used (the initial condition cannot be used in this case, x < t). For x < t,
du dt
= x = t − η or η = t − x.
Integrating gives u(x, t) = 12 t2 − ηt + B. Using the boundary condition, we find B = u(x, t) − 12 t2 + ηt = u(0, η) − 12 η 2 + η 2 = η + 12 η 2 .
6.12 Exercises
799
With η = t − x, the solution is given by u(x, t) = 12 t2 − (t − x)t + (t − x) + 12 (t − x)2 = 12 x2 + t − x. On x = t, u(x, t) = 21. We have
dx x2
=
dt u
t2 . 2
=
du 1
= dz gives
dx dz
= x2 ,
du dz
= 1,
dt dz
= u, whence
x = (a − z)−1 , u = z + b, t = 12 z 2 + bz + c, when a, b, and c are constants. The characteristic equation originating from the initial curve at x = s, t = 1 − s ), u = z(b = 0), t = 12 z 2 + 1 − s. s(z = 0) is x = ( 1−sz
For each x, t, the solution u depends on the parameters s and z. Eliminating s and z gives the answer as follows: s = s=
x 1+zx
1 2 2z
+ 1 − t = ( 12 u2 + 1 − t) and
x = ( 1+ux ). Thus, ( 12 u2 + 1 − t)(1 + ux) = x.
22. The Darcy law for water is Vd = −KHx , K is the hydraulic conductivity and H is the hydraulic head, H =
p gρ
+z+
v2 2g
≈
p gρ
+ z for small (v 2 /2g), p is
the pressure, g is the acceleration to gravity, and ρ is the density. The Darcy law describes the assumption that water flows slowly in the soil pores for the head loss to be proportional to the velocity. The characteristic equations associated with Buckley–Leverett equation are dt 1
=
dx c
=
ds . 0
Or equivalently,
conserved quantity. Hence,
dx dt
ds dt
= 0, and hence, s = const. and it is a
= c is the speed at which the interface between
water and hydrocarbon moves. Thus, c =
∂F ∂s
= 0 when s = 0 and s = 1, and c
attains its maximum when 0 < s < 1. The case c = 0 for s = 0 reflects a well known property of hydrocarbons as non-wetting liquids, the mobility of which is very low for small water saturation.
6.12 Exercises 1. Hint: Use ht + (uh)x = 0.
800
C Answers and Hints to Selected Exercises
⎧ ⎪ ⎪ 200 ⎪ ⎪ ⎪ ⎨ 2. ρ(x, t) = f (ξ) = 100(1 − ξε ) ⎪ ⎪ ⎪ ⎪ ⎪ ⎩0
if ξ < −ε, if − ε < ξ < ε, if ξ > ε,
⎧ ⎪ ⎪ ξ − 60t if ξ < −ε, ⎪ ⎪ ⎪ ⎨ x = ξ + [60 − 35 f (ξ)]t = (60t − ε)( ξε ) if − ε < ξ < ε, ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ ξ + 60t if ξ > ε, ⎧ ⎪ ⎨ 200 if x < −(60t + ε), ρ(x, t) = ⎪ ⎩ 100(1 − x ) if − (60t + ε) < x < 60t + ε. 60t+ε The graph of ρ(x, 0) is linearly increasing in x from −ε to +ε. ⎧ ⎪ ⎨ 25(3 − |ξ|) if |ξ| ≤ 1, and 3. (a) ρ = f (ξ) = ⎪ ⎩ 50 if |ξ| ≥ 1, ⎧ ⎪ ⎨ ξ = 15(1 + |ξ|)t if |ξ| ≤ 1, x= ⎪ ⎩ ξ + 30t if |ξ| ≥ 1. 1 , we eliminate ξ from the above solution to obtain (b) For t < 15 ⎧ ⎪ ⎪ 25(3 + x−15t ⎪ 1−15t ) if − 1 + 30t ≤ x ≤ 15t, ⎪ ⎨ ρ(x, t) = 25(3 − x−15t ) if 15t ≤ x ≤ 1 + 30t, ⎪ x+15t ⎪ ⎪ ⎪ ⎩ 50 if |x − 30t| ≥ 1.
The solution ρ(x, t) for t =
1 (1 15
− ε) can be found from the result given in
Exercise 3(a). Use results in 3(a) to find ρ and x for t = ⎧ ⎪ ⎨ 25(3 − |ξ|) if |ξ| ≤ 1, ρ= and ⎪ ⎩ 50 if |ξ > 1|,
1 15 .
6.12 Exercises
⎧ ⎪ ⎪ 2+ξ ⎪ ⎪ ⎪ ⎨ x = 1 + 2ξ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩1
801
if ξ > 1orξ < −1, if 0 ≤ ξ ≤ 1, if − 1 ≤ ξ < 0.
1 ). Use these results to draw a graph of ρ(x, 15
4. Hint: Use u = x − c0 t and v = t as independent variables. 12. Hint: The equation in matrix form is ⎛ ⎜u ⎜ ⎜1 ⎜ρ ⎝ 0
c2 ρ u 0
⎞⎛ ⎞ ⎛ 0 ⎟ ⎜ px ⎟ ⎜ 1 ⎟⎜ ⎟ ⎜ ⎜ ⎟ ⎜ 0⎟ ⎟ ⎜ ux ⎟ + ⎜ 0 ⎠⎝ ⎠ ⎝ Sx 0 u
⎞⎛ ⎞ 0 0 ⎟ ⎜ pt ⎟ ⎟⎜ ⎟ ⎜ ⎟ 1 0⎟ ⎟ ⎜ ut ⎟ = 0. ⎠⎝ ⎠ St 0 1
The eigenvalues are the roots of the equation u − λ c2 ρ 1 u−λ ρ 0 0
dx = u, u ± c. 0 = 0 which gives λ = dt u − λ 0
13. Hint: du + λ2,1 dv = 0 along the characteristics C1 and C2 , respectively. a(du)2 + 2b du dv + c(dv)2 = 0, which corresponds to the characteristics dx = λ dt, a(dx)2 − 2b dx dt + c(dt)2 = 0. 14. Hint: Along the characteristics C+ : η = x − ct = const. and
du+ dt
+ 12 au+ = 0
with the solution u+ (t) = u+ (0) exp(− 12 at) with η = x − ct and u− (t) = u− (0) exp(− 12 at) with ξ = x + ct. 15. Hint: ρ = ρ(x(t), t),
dρ dt
=
∂ρ ∂t
+
dx dt
·
∂ρ ∂x ,
and then use equations (6.3.1) and
(6.3.2). 16. Hint: λ1,2 = u ± c, L1,2 = (1, ±2). ∂ ∂ + λ1 ∂x )(u + 2c) = Hx along C+ : ( ∂t
dx dt
= u + c,
∂ ∂ + λ2 ∂x )(u − 2c) = Hx along C− : ( ∂t
dx dt
= u − c.
802
C Answers and Hints to Selected Exercises
18. Hint: See Jeffrey and Taniuti (1964, pp. 72, 73). 19. Hint: Equate xsr and xrs to derive the Euler–Poisson–Darboux equation. 22. Hint: The first equation can be written as
∂ρ ∂t
+ div(ρu, ρv, ρw) = 0.
The first and second equations can be combined to find the result ⎛ ⎞ 2 ρu ⎜ p + ρu ⎟ ⎜ ⎜ ⎟ ∂ ⎜ ⎜ ρv ⎟ + div⎜ ρuv ⎜ ⎟ ⎜ ∂t ⎝ ⎝ ⎠ ρw ρuw ⎛
ρuv p + ρv 2 ρvw
⎞ ρuw ⎟ ⎟ ρvw ⎟ ⎟ = 0. ⎠ p + ρw2
The above two results lead to the desired conservation form. ⎛ ⎞ ⎛ ⎞ c u 0 0 a a ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟. 25. For (i), A = ⎜ −1 0 0 ⎟ and B = ⎜ 0 ⎟ ⎝ ⎠ ⎝ ⎠ 0 0 1 −v − w ⎛ ⎛ ⎞ ⎞ c u c 0 − a a a ⎜ ⎜ ⎟ ⎟ ⎜ ⎜ ⎟ ⎟ ⎜ ⎜ ⎟ ⎟, c u c For (ii), A = ⎜−1 − a a 0 ⎟ and B = ⎜ ⎟ a ⎝ ⎝ ⎠ ⎠ 1 v w −2 − 2 0 0 2 and thus, for both cases (i) and (ii), ⎞ v ⎜ ⎟ ⎜ ⎟ ⎟ U =⎜ ⎜w⎟. ⎝ ⎠ u ⎛
26. Characteristics drawn from the negative x-axis are vertical lines x = const., and along these characteristics, ρ = ρ0 and ρ(x, t) = ρ0 for x < 0, t > 0. Characteristics drawn from points x > a satisfy x(t) ˙ = c(ρ(x, 0), 0) = c(0) = ρm . If x > a + ρm t, t > 0, ρ(x, t) = 0. If 0 < x < a, ρ(x, t) = f (x − tc(ρ)) = ρ0 [a − {x − tρm (1 − Hence, x = (a −
ρ ρ0 )
+ ρm (a −
ρ ρ ρ0 )t exp(− ρ0 ).
ρ ρ ρ0 ) exp(− ρ0 )}].
7.9 Exercises
803
27. (a) This system of two equations is known to be hyperbolic. The first equation is already in the one-dimensional divergence form because it becomes ρt + (ρu)x = 0. The one-dimensional form of the second equation is not in divergence form, but it is multiplied by ρ and added to u times the first equation to obtain the conservation form (ρu)t + (ρu2 + p)x = 0. (c) We have ρt + (ρu)x = 0, or equivalently, Thus,
dρ du
·
∂u ∂t
equivalently, Thus,
∂u ∂t
∂ρ ∂u ∂u ∂t
dρ ∂u + (ρ + u du )( ∂x ) = 0. It follows that ∂u ∂t
+ (uux +
+ (u +
1 ∂p ρ ∂x )
2
c (ρ) dρ ∂u ρ du ) ∂x
+ (ρx u + ρux ) = 0. ∂u ∂t
+ (uux ) +
1 ∂p ρ ∂x
= 0, or
= 0.
= 0.
28. We have xvt +wx = 0 and close the system by using vx = wt . Thus, the Tricomi equation becomes Ut + AUx = 0, where U =
v w
eigenvalues λ2 =
− x1
and A =
0
1 x
. This has
−1 0
for x = 0.
When x < 0, the eigenvalue are real and distinct, and the system is hyperbolic. When x > 0, the system is elliptic, and the system is parabolic when x = 0. 29. The first equation is in conservation form: ht + (uh)x = 0. Multiply the second equation by h and add the result to u times the first equation to obtain the conservation form.
7.9 Exercises 5. Hint:
u2 ˜ xut dx = − x ∂x + Ku dx 2 −∞ −∞ ∞ ∞ ∞ 1 2 1 ˜ u + Ku dx = I2 (u) + dx K(x − s)u(s, t) ds = 2 −∞ 2 −∞ −∞ 1 = I2 (u) + c(0)I1 (u). 2
I˙4 =
∞
∞
Integrating with respect to t gives I4 (u) − 12 t I2 (u) − tc(0)I1 (u) = const.
804
C Answers and Hints to Selected Exercises
6. Hint: (a) Multiply the given equation by
√ t to get
√ √ tu t + t uxx − 3u2 x = 0. (b) Multiply the given equation by 2tu to obtain
tu2
t
+ t 2uuxx − u2x − 4u3 x = 0.
7. Hint: Multiply equation (2.7.67) by u and follow the procedure described by Benney (1974) and Miura (1974) to obtain
1 3 u 3
+
t
1 4 u 4
+ wu2 uz + u2 hx = 0, x
or equivalently, ( 13 u3 )t + ( 14 u4 + u2 h)x + ( 13 u3 w)z − 13 u3 wz − 2huux = 0. Multiply equation (2.7.67) by h to obtain (hu)t − uht + huuxx + hwuz +
1 2 h x = 0. 2
This equation is added to the previous equation to generate 1 3 1 4 1 2 1 2 hu + u + u h+ h + h + u3 w z − huux 3 4 2 3 t x + hwuz − huux − uht = 0.
Write m =
h 0
u dz so that ht + mx = 0, and then 1 3 1 4 1 2 2 hu + u + hu + um + u + h 3 3 2 x x 1 3 + w hu + m + u = 0. 3 z
This gives the answer.
8.14 Exercises
8. Hint: (a) (
∞ −∞
u dx)y = 0, and hence,
∞ −∞
805
u dx = c(t).
Under suitable conditions, c(t) is a constant, and hence, it can be set equal to zero. To prove (c), we find
∞
−∞
∞ u dx + uxx − 3u2 −∞ + 3 t
∞
v dx −∞
= 0. y
Both (b) and (c) represent the conservation of momentum in both x- and ydirections. 10. It follows from the definition of H that
∂H ∂J
= ω, which means that the frequency
of the modulated wave is determined by the partial derivative of H with respect to the wave action J. Thus, ∂k ∂t
+
∂J ∂t
+
∂ ∂H ( ) ∂x ∂J ∂ ∂H ∂x ( ∂k )
∂k ∂t
+
∂ω ∂x
= 0 reduces to the Hamiltonian form
= 0. We next replace Lω by J in the Whitham equation to obtain = 0 because
∂H ∂k
= − ∂L ∂k . Thus, the above equation describes
represents the density of its flux. conservation of the wave action J, and ∂H ∂k ∞ The quantity I = −∞ J dx is conserved in a localized wave packet. It is noted that variables J and k are introduced to describe the modulated wave, and they play the role similar to the canonical variables of the action-angle in classical mechanics.
8.14 Exercises 1. (a) α = β = 1 and γ = 2, (b) β = 2α, (c) γ = α − β, (d) β = 2α. 2. In D1 and for c ≥ 2, u(ξ) = A exp(m1 ξ) + B exp(m2 ξ), where m1 , m2 = √ 1 c2 − 4]. In D2 , u(ξ) = 1 + A1 exp(n1 ξ) + B1 exp(n2 ξ), where n1 , 2 [−c1 ± √ n2 = 12 [−c ± c2 + 4]. To determine A, B, A1 , and B1 , we use the facts that (i) u(0) = 12 , (ii) the values of u (0) in D1 and D2 are equal, and (iii) u(−∞) = 1 and u(∞) = 0. In D1 ,
806
C Answers and Hints to Selected Exercises
for ξ > 0, we have 1 c2 − 4 + 2c − c2 + 4 exp(m1 ξ) u(ξ) = √ 2 4 c− 4 c2 − 4 − 2c + c2 + 4 exp(m2 ξ) . + In D2 , for ξ < 0, we find u(ξ) = 1 −
1 2
exp(n1 ξ).
When c = 2, we find solutions in D1 and in D2 . In D1 , for ξ > 0, u(ξ) = 12 exp(−ξ)+(1− √12 )ξ exp(−ξ), and in D2 , for ξ < 0, √ u(ξ) = 1 − 12 exp[( 2 − 1)ξ]. 8. Hint: ψ˜ = aβ+1 ψ, and set β = 0 to find the similarity solutions. 9. Hint: The boundary conditions are u(η = 0) = 0 and u(η → ∞) = 1, v(η = 0) = vω . 11. v +
c2 2 4 (η
v + 5 η v + 3v) = 0.
) η , ut = Hint: ηt = −( α β t
γ
t β −1 (γv − α η v ),
1 β
2γ α 2 2 utt = t β −2 {( α β ) η v + β (− β + γ
α β
+ 1)η v + ( βγ − 1)( βγ )v},
ux = t β − 2 v , uxxxx = t β −2 v . γ
γ
1
13. Hint:
a
˙ Q(t) =2
ux uxt dx =
2[ux ut ]a0
a
−2
0
uxx ut dx 0
uxx uxx + F (u) dx 0 a a = −2 u2xx dx + 2 u2x F (u) dx. a
= −2
0
0
Using the boundary condition gives the Poincaré inequality
a
u2xx 0
and
dx ≥ π
a
2
u2x dx 0
9.14 Exercises
a
807
a u2x F (u) dx ≤ A u2x dx, 0 0 ˙ Q(t) ≤ 2 A − π 2 Q(t) = μQ(t).
u2x F (u) dx ≤
0
a
Integrating from 0 to a gives the inequality. 14. See Murray (1997), p. 357. 20. Combining ut + (ρu)x = 0 with (ρu)t + (ρu2 )x = 0 gives ut + uux = 0. Or in other words, ut + ( 12 u2 )x = 0. The characteristic equations are dx u
=
du 0 ,
which leads to u = const. for
dx dt
dt 1
=
= u. Thus, u is invariant, and ρ
and ρu are conserved quantities. Writing U = u and F =
1 2 2u
=
1 2 2U
gives
Ut + Fx = 0. Or equivalently, Ut + FU Ux = 0, that is, Ut + cUx = 0, where c =
∂F ∂U
= U = u is the propagation velocity. Differentiating the inviscid
Burgers equation ut + uux = 0 with respect to x gives (ux )t + u(ux )x = −u2x . Thus, the characteristic equations for ux are d u dt x
= −u2x for
dx dt
dt 1
=
dx u
=
dux , −u2x
that is,
= u. Integrating gives − u1x = −t + A, where A is a
constant. When t = t0 , (ux ) = u0 , and A = t0 − u1 . Thus, the solution becomes 0
(ux )(t) =
u0 [1
+ (t −
t0 )u0 ]−1 .
9.14 Exercises 20. Hint: Add two equations in 20(d) and integrate the result to show
dE dt
= 0.
21. Hint: If u0 (x) = (c1 − γ) exp(−|x − x1 |) + (c2 − γ) exp(−|x − x2 |), x ∈ R with c1 < γ, c2 > γ and x1 < x2 , verify that m0 = u0 − u0xx = 2c1 δ(x − x1 ) + 2c2 δ(x − x2 ),
where δ(x) is the Dirac delta function. 22. Hint: u(x, t) is a solution of
808
C Answers and Hints to Selected Exercises
T
0
R
(uφt + F (u)φx ) dx dt +
u(x, 0)φ(x, 0) dx = 0, R
for any T > 0 and φ ∈ C0∞ ([0, T ] × R) with F (u) = 12 u2 + p ∗ ( 32 u2 ) + γu, where p(x) is defined in 21(a). (See Guo 2009.)
10.9 Exercises 10. We seek the solution ψ(x, t) = Ψ (x)e−iωt where ω = (E/) is the frequency of the De Broglie wave and the function Ψ (x) satisfies the equation Ψ (x) +
2mE (1 − x)Ψ = 0. 2 1
Making the change of variable z = −(1 − x)(2mE/2 ) 3 reduces the equation into the form Ψ − zΨ = 0. The bounded solution at |z| → ∞ can be expressed in terms of the Airy function Ψ (z) = CAi (z) = C 0
∞
s3 ds, cos sz + 3
where C is a constant. Thus, the solution describing the behavior of the electron is ψ(z, t) = CAi (z) exp(−iωt). The function |ψ|2 describes the probability of the electron at a point z. For z > 0, the probability drops to zero suddenly as if there were an obstacle reflecting the electron at z = 0 which is referred to as the turning point. For z < 0, the solution ψ(z, t) represents a nonuniform standing wave. Using the asymptotic expansion of Ai (z) for |z| → ∞, in the form
Ai (z) ∼
⎧ ⎪ 1 ⎪ ⎨√
1
exp(− 23 z 3/2 ),
when z → ∞,
⎪ ⎪ ⎩√1
1
cos( 23 |z|3/2 + π4 ),
when z → −∞.
πz 4
πz 4
12.10 Exercises
809
The solution for ψ(z, t) can be written as the superposition of two progressive waves in the form
2 3/2 π C exp −i ωt + |z| − ψ(z, t) ∼ √ 4 z 3 4 π 2 . + exp −i ωt − |z|3/2 + 3 4 This represents the De Broglie wave for the particle in free space.
11.14 Exercises 6. Hint: φ(x, t) = tan( u4 ),
x − Ut x + Ut ∓ U exp ∓ √ φ(x, t) ∼ ±U exp ± √ 1 − U2 1 − U2 Note that ∓U exp(θ) = ± exp{θ +
1√ γ 2 1−U 2 },
where
γ 2
as t → ∓∞.
1
= 1(1 − U ) 2 log U .
+a2 ) tan{ 14 (u2 − u3 )}] for any solutions u2 , u3 . 14. (b) u4 = 4 tan−1 [( aa11 −a 2
ur tan 4
= exp(θr ),
θr = (ar x + a−1 r t + εr ), r = 2, 3.
15. Hint: Use ux (x, t) instead of u(x, t). 22. vtt − c2 vxx − a2 v = 0. 1 24. u(x, t) = 2c J0 [d t2 −
x2 c2 ]H(ct
− |x|).
12.10 Exercises 1. (a) The continuity and the momentum equations for the inviscid Burgers equa tion can be written in the matrix form Ut + AFx = 0 by defining U = ρ = ρu
u1 u2
,F =
Thus, A =
F1 F2
∂F1 ∂u1 ∂F2 ∂u1
=
ρu
2
ρu ∂F1 ∂u2 ∂F2 ∂u2
=
=
u2 u22 /u1
0 u −( u2 )2 1
1
2u2 u1
.
=
0 1 −u2 2u
.
810
C Answers and Hints to Selected Exercises
The eigenvalues of A are the roots of 0 = |A − λI| =
−λ 1 −u2 2u−λ
= λ2 −
2uλ + u2 . Thus, A has two equal eigenvalues (λ = u, u) which are not distinct. Hence, the system is not hyperbolic. ∂F 1 ∂F1 ∂u1 ∂u2 = ∂P 0 u2 2 2u12 = (b) A = ∂F ∂F 2 ∂u1
2 ∂u2
∂u1
−( u )
u1
1
0 1 c2 −u2 2u
, where c =
∂P ∂u1 .
The
eigenvalues are real and distinct roots of |A − λI| = 0. Or, |A − λI| = 2−λ 2 1 = 0, that is, λ = λ1 , λ2 = u ∓ c. c −u 2u−λ
The corresponding eigenvectors are be defined by B =
1 1 u−c u+c
diagonal matrix, D =
1 u−c
and
1 u+c
, and hence, B −1 =
u−c 0 0 u+c
=
λ1 0 0 λ2
, and a new matrix B can
1 c+u −1 2c c−u 1
with B −1 AB =
.
u The Riemann invariants are dR = B −1 dU , and hence, dR1 = ( 12 + 2c )dU1 − 1 1 u 1 R1 . 2 dU2 and dR2 = ( 2 − 2c )dU1 + 2 dU2 , where the matrix R = R2
(c) The 2 × 2 matrix system can be generalized for an n × n system as Ut + Fx = S. Or equivalently, Ut + AUx = S, where U , F , and S are n × 1 matrices given by ⎤
⎡
⎤
⎡ F1
u1
⎢ ⎥ ⎢ ⎥ ⎢ u2 ⎥ ⎢ ⎥ U = ⎢ . ⎥, ⎢ . ⎥ ⎢ . ⎥ ⎣ ⎦ un
⎢ ⎥ ⎢ ⎥ ⎢ F2 ⎥ ⎢ ⎥ F = ⎢ . ⎥, ⎢ . ⎥ ⎢ . ⎥ ⎣ ⎦ Fn
S1
⎢ ⎥ ⎢ ⎥ ⎢ S2 ⎥ ⎢ ⎥ S = ⎢ . ⎥, ⎢ . ⎥ ⎢ . ⎥ ⎣ ⎦ Sn
A is the n × n Jacobian matrix of F : ⎤
⎡ ∂F1 ∂u1
⎢ ⎢ ⎢ ∂F2 ⎢ ∂u 1 A=⎢ ⎢ . ⎢ .. ⎢ ⎣
∂Fn ∂u1
∂F1 ∂u2
···
∂F2 ∂u2
.. .
··· .. .
∂Fn ∂u2
···
⎤
⎡
∂F1 ∂un
⎥ ⎥ ∂F2 ⎥ ⎥ ∂un ⎥ . .. ⎥ ⎥ . ⎥ ⎦
∂Fn ∂un
12.10 Exercises
811
A hyperbolic system of conservation laws is a system of conservation laws with the following properties: (i) The components of F and S are functions of components of U and possibly of x and t, but F does not contain any derivative of U with respect to x and t, and (ii) the matrix A has n real and distinct eigenvalues which are roots of |A − λI| = 0. The theory of hyperbolic systems of conservation laws with many examples of applications is available in many books including Defermos (2000), Holden and Risebro (2002), Jeffrey (1976), Lax (1973, 2006), Serre (2000), Sharma (2010), and Zheng (2001).
Bibliography1
Abdelkader, M.A. (1982). Travelling wave solutions for a generalized Fisher equation, J. Math. Anal. Appl. 85, 287–290. Ablowitz, M.J. and Clarkson, P.A. (1991). Solitons, Nonlinear Evolution Equations and Inverse Scattering, Cambridge University Press, Cambridge. Ablowitz, M.J. and Haberman, R. (1975). Nonlinear evolution equations-two and three dimensions, Phys. Rev. Lett. 35, 1185–1188. Ablowitz, M.J. and Haut, T.S. (2009). Asymptotic expansions for solitary gravitycapillary waves in two and three dimensions, Proc. R. Soc. Lond. A465, 2725– 2749. Ablowitz, M.J. and Haut, T.S. (2010). Asymptotic expansions for solitary gravitycapillary waves in two and three dimensions, J. Phys. A, Math. Theor. 43(43), 1–14. #434005 Ablowitz, M.J. and Ladik, J.F. (1976a). Nonlinear differential-difference equations and Fourier analysis, J. Math. Phys. 17, 1011–1018. Ablowitz, M.J. and Ladik, J.F. (1976b). A nonlinear difference scheme and inverse scattering, Stud. Appl. Math. 55, 213–229. Ablowitz, M.J. and Segur, H. (1979). On the evolution of packets of water waves, J. Fluid Mech. 92, 691–715. Ablowitz, M.J. and Segur, H. (1981). Solitons and the Inverse Scattering Transforms, SIAM, Philadelphia. Abramowitz, M. and Stegun, I.A. (1972). Handbook of Mathematical Functions, Dover, New York. Ablowitz, M.J. and Zeppetella, A. (1979). Explicit solutions of Fisher’s equation for a special wave speed, Bull. Math. Biol. 41, 835–840. Ablowitz, M.J., Kaup, D.J., Newell, A.C., and Segur, H. (1973). Method for solving the sine-Gordon equation, Phys. Rev. Lett. 30, 1262–1264. 1
This bibliography is not, by any means, a complete one for the subject. For the most part, it consists of books and papers to which reference is made in the text. Many other selected books and papers related to material in this book have been included so that they may stimulate new interest in a future study and research.
L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1,
814
Bibliography
Ablowitz, M.J., Kaup, D.J., Newell, A.C., and Segur, H. (1974). The inverse scattering transform Fourier analysis for nonlinear problems, Stud. Appl. Math. 53, 249–315. Airy, G.B., (1845). Tides and Waves, in Encyclopedia Metropolitana, Article 192. Akylas, T.R. (1993). Envelope solitons with stationary crests, Phys. Fluids 5, 789– 791. Akylas, T.R. (1994). Three-dimensional long water-wave phenomena, Annu. Rev. Fluid Mech. 26, 191–210. Akylas, T.R. (2007). Solitary waves in rotating fluids, in Solitary Waves in Fluids (ed. R.H.J. Grimshaw), 111–124. Akylas, T.R. and Grimshaw, R. (1992). Solitary internal waves with oscillatory tails, J. Fluid Mech. 242, 279–298. Akylas, T.R., Dias, F., and Grimshaw, R. (1998). The effect of the induced mean flow on solitary waves in deep water, J. Fluid Mech. 355, 317–328. Ames, W.F. (1965). Nonlinear Partial Differential Equations in Engineering, Vol. I, Academic Press, New York. Ames, W.F. (1972). Nonlinear Partial Differential Equations in Engineering, Vol. II, Academic Press, New York. Amick, C. and Kirchgassner, K. (1989). A theory of solitary waves in the presence of surface tension, Arch. Ration. Mech. Anal. 105, 1–49. Amick, C.J., Bona, J.L., and Schonbek, M.E. (1989). Decay of solutions of some nonlinear wave equations, J. Differ. Equ. 81, 1–49. Amick, C.J., Fraenkel, L.E., and Toland, J.F. (1982). On the Stokes Conjecture for the wave of extreme form, Acta Math. Stockh. 148, 193–214. Ammerman, A.J. and Cavalli-Sforva, L.L. (1971). Measuring the rate of spread of early farming, Man 6, 674–688. Ammerman, A.J. and Cavalli-Sforva, L.L. (1983). The Neolithic Transition and the Genetics of Populations in Europe, Princeton University Press, Princeton. Aoki, K. (1987). Gene-culture waves of advance, J. Math. Biol. 25, 453–464. Arecchi, F.T., Masserini, G.L., and Schwendimann, P. (1969). Electromagnetic propagation in a resonant medium, Rev. Nuovo Cimento 1, 181–192. Aris, R. (1975). The Mathematical Theory of Diffusion and Reaction in Permeable Catalyst, Oxford University Press, Oxford. Arnold, R., Showalter, K., and Tyson, J.J. (1987). Propagation of chemical reactions in space, J. Chem. Educ. 64, 740–742. Aronson, D.G. (1980). Density-dependent interaction–diffusion systems, in Dynamics and Modelling of Reactive Systems (eds. W.E. Stewart, W.H. Ray, and C.C. Conley). Academic Press, Boston, 161–176. Aronson, D.G. (1986). The porous medium equation, in Nonlinear Diffusion Problems, Lecture Notes in Mathematics, Vol. 1224, Springer, New York, 1–46. Asano, N. (1970). Reductive perturbation method for nonlinear wave propagation in inhomogeneous media III, J. Phys. Soc. Jpn. 29, 220–224. Asano, N. (1974). Modulation for nonlinear wave in dissipative or unstable media, J. Phys. Soc. Jpn. 36, 861–868. Asano, N. and Ono, H. (1971). Nonlinear dispersive or dissipative waves, J. Phys. Soc. Jpn. 31, 1830–1836.
Bibliography
815
Asano, N. and Taniuti, T. (1970). Reductive perturbation method for nonlinear wave propagation in inhomogeneous media II, J. Phys. Soc. Jpn. 29, 209–214. Asano, N., Taniuti, T., and Yajima, N. (1969). A perturbation method for a nonlinear wave modulation II, J. Math. Phys. 10, 2020–2024. Atiyah, M., Gibbon, J.D., and Wilson, G. (1985). New developments in the theory and applications of solitons, Philos. Trans. R. Soc. Lond. 315, 333–469. Avilov, V.V., Krichever, I.M., and Novikov, S.P. (1987). Evolution of the Whitham zone in the Korteweg–de Vries Theory, Sov. Phys. Dokl. 32, 345–349. Balmforth, N.J. (1995). Solitary waves and homoclinic orbits, Annu. Rev. Fluid Mech. 27, 335–373. Barcilon, A. and Drazin, P.G. (2001). Nonlinear waves of vorticity, Stud. Appl. Math. 106(4), 437–479. Barenblatt, G.I. (1979). Similarity, Self-similarity and Intermediate Asymptotics, Consultants Bureau, New York. Barenblatt, G.I. and Zel’dovich, Y.B. (1972). Self-similar solutions as intermediate asymptotics, Annu. Rev. Fluid Mech. 4, 285–312. Barone, A., Esposito, F., Magee, C.J., and Scott, A.C. (1971). Theory and applications of sine-Gordon equation, Rev. Nuovo Cimento 1(2), 227–267. Bebernes, J. and Eberly, D. (1989). Mathematical Problems from Combustion Theory, Springer, New York. Benguria, R. and Depassier, M. (1989). Equations of the Korteweg–de Vries type with nontrivial conserved quantities, J. Phys. A 22, 4135–4142. Benjamin, T.B. (1962). The solitary wave on a stream with arbitrary distribution of vorticity, J. Fluid Mech. 12, 97–116. Benjamin, T.B. (1967). Instability of periodic wave trains in nonlinear dispersive systems, Proc. R. Soc. Lond. A299, 59–75. Benjamin, T.B. (1992). A new kind of solitary wave, J. Fluid Mech. 245, 401–411. Benjamin, T.B. and Feir, J.E. (1967). The disintegration of wavetrains on deep water, Part 1, Theory, J. Fluid Mech. 27, 417–430. Benjamin, T.B. and Lighthill, M.J. (1954). On cnoidal waves and bores, Proc. R. Soc. Lond. A224, 448–460. Benjamin, T.B. and Olver, P.J. (1982). Hamiltonian structure, symmetrics and conservation laws for water waves, J. Fluid Mech. 125, 137–185. Benjamin, T.B., Bona, J.L., and Mahony, J.J. (1972). Model equations for long waves in nonlinear dispersive systems, Philos. Trans. R. Soc. A272, 47–78. Benney, D.J. (1966). Long waves on liquid films, J. Math. Phys. 45, 150–155. Benney, D.J. (1974). Long waves, in Nonlinear Wave Motion (ed. A.C. Newell), Am. Math. Soc., Providence, 49–60. Benney, D.J. and Roskes, G. (1969). Wave instabilities, Stud. Appl. Math. 48, 377– 385. Ben-Jacob, E., Brand, H., Dee, G., Kramer, L., and Langer, J.S. (1985). Pattern propagation in nonlinear dissipative systems, Physica D 14, 348–364. Benton, E.R. and Platzman, G.W. (1972). A table of solutions of one-dimensional Burgers equation, Q. Appl. Math. 30, 195–212. Berezin, Y.A. and Karpman, V.I. (1966). Nonlinear evolution of disturbances in plasmas and other dispersive media, Sov. Phys. JETP 24, 1049–1056.
816
Bibliography
Bespalov, V.I. and Talanov, V.I. (1966). Filamentary structure of light beams in nonlinear liquids, JETP Lett. 3, 307–310. Bhatnagar, P.L. (1979). Nonlinear Waves in One-Dimensional Dispersive Systems, Oxford University Press, Oxford. Bhatnagar, P.L. and Prasad, P. (1970). Study of the self-similar and steady flows near singularities, Part I, Proc. Roy Soc. London A315, 569–584. Bhatnagar, P.L. and Prasad, P. (1971). Study of the self-similar and steady flows near singularities, Part II, Proc. Roy Soc. London A322, 45–62. Bhatta, D. and Debnath, L. (2003). Small amplitude wave in shallow water over linear and quadratic sloping beds, J. Appl. Math. Comput. 13, 53–65. Biodini, G. and Kodama, Y. (2003). On a family of solutions of the Kadomtsev– Petviashvili equation which also satisfy the Toda lattice hierarchy, J. Phys. A, Math. Gen. 36, 10519–10536. Biondini, G. and Wang, D. (2010). On the soliton solutions of the two-dimensional Toda Lattice, J. Phys. A, Math. Theor. 43(43), 1–20. #43007 Birkhoff, G. (1950). Hydrodynamics, Princeton University Press,Princeton. Birkhoff, G.D. (1927). Dynamical Systems, Am. Math. Soc., Providence. Blackstock, D.T. (1964). Thermoviscous attenuation of plane, periodic finiteamplitude sound waves, J. Acoust. Soc. Am. 36, 534–542. Bluman, G.W. and Kumei, S. (1989). Symmetries and Differential Equations, Springer, New York. Bodini, G. (2007). Line soliton interactions of the Kadomtsev–Petviashvili equation, Phys. Rev. Lett. 99, 064103. Bodini, G., Maruno, K., Oikawa, M., and Tsuji, H. (2009). Soliton interactions of the Kadomtsev–Petviashvili equation and generation of large-amplitude waves, Stud. Appl. Math. 122, 377–394. Bogdanov, L.V. (2010). Non-Hamiltonian generalizations of the dispersionless 2DT L hierarchy, J. Phys. A, Math. Theory 43, 1–8, #434008. Bogoyavlensky, O.I. (1976). On perturbations of the periodic Toda Lattice, Commun. Math. Phys. 51, 201–209. Bona, J.L. and Schonbek, M.E. (1985). Travelling-wave solution to the Korteweg– de Vries–Burgers equation, Proc. R. Soc. Edinb., Sect. A 101, 207–226. Born, M. and Infeld, L. (1934). Foundations of a new field theory, Proc. R. Soc. Lond. A144, 425–451. Bouchut, F., Le Sommer, J., and Zeitlin, V. (2005). Breaking of balanced and unbalanced equatorial waves, Chaos 15(1), 013503. Boussinesq, J. (1871a). Théorie de l’intumescene liquide appelée onde solitaire ou de translation se propageant dans un canal rectangulaire, Comptes Rendus 72, 755–759. Boussinesq, M.J. (1871b). Théorie generale des mouvements qui sont propagés dans un canal rectangulaire horizontal, Comptes Rendus 72, 256–260. Boussinesq, M.J. (1872). Théorie des ondes et des rumous qui se propagent le long d’un canal rectagulaire horizontal, en communiquant au liquide contenu dans ce canal des vitesses sensiblement pareilles de la surface au fond, J. Math. Pures Appl. Ser. (2) 17, 55–108.
Bibliography
817
Boussinesq, M.J. (1877). Essai sur la théorie des eaux courants. Mémoires présentes par diviers savants à l’Académie des Sciences Inst. France (Series 2) 23, 1–630. Boyd, J.P. (1980). The nonlinear equatorial Kelvin wave, J. Phys. Oceanogr. 10, 1–11. Boyd, J.P. (1983). Equatorial solitary waves. Part 2: Envelope solitons, J. Phys. Oceanogr. 13, 428–449. Boyd, J.P. (2007). Planetary solitary waves, in Solitary Waves in Fluids (ed. R.H.J. Grimshaw), 125–158. Boyd, J.P. and Ma, H. (1990). Numerical study of elliptical modons by a spectral method, J. Fluid Mech. 221, 597–611. Bramson, M. (1983). Convergence of solutions of the Kolmogorov equation to travelling waves, Mem. Am. Math. Soc. 285, 1–190. Bressan, A. (2000). Hyperbolic Systems of Conservation Laws. The OneDimensional Cauchy Problem, Oxford University Press, Oxford. Bressan, A. (2007). Global dissipative solutions of the Camassa–Holm equation, Anal. Appl. 5, 1–27. Bressan, A. and Constantin, A. (2007). Global conservative solutions of the Camassa–Holm equation, Arch. Ration. Mech. Anal. 183, 215–239. Britton, N.F. (1986). Reaction-Diffusion Equations and Their Applications to Biology, Academic Press, New York. Buckley, S.E. and Leverett, M.C. (1942). Mechanisms of fluid displacement in sands, Trans. AIME 146, 107–116. Burgers, J.M. (1948). A mathematical model illustrating the theory of turbulence, Adv. Appl. Mech. 1, 171–199. Burgers, J.M. (1974). The Nonlinear Diffusion Equation, Reidel, Dordrecht. Burton, C.V. (1893). On plane and spherical sound waves of finite amplitude, Philos. Mag. 35, 317–333. Calogero, F. and Degasperis, A. (1978). Solution by the spectral-transform method of a nonlinear evolution equation including as a special case the cylindrical KdV equation, Lett. Nuovo Cimento 19, 150–154. Calogero, F. and Degasperis, A. (2005). Novel solution of the system describing the resonant interaction of three waves, Physica D 200, 242–256. Calvo, D.C. and Akylas, T.R. (2002). Stability of steep gravity-capillary solitary waves in deep water, J. Fluid Mech. 452, 123–143. Calvo, D.C., Yang, T.S., and Akylas, T.R. (2000). On the stability of solitary waves with decaying oscillatory tails, Proc. R. Soc. 456, 469–487. Camassa, R. and Holm, D.D. (1993). An integrable shallow water equation with peaked solitons, Phys. Rev. Lett. 71, 1661–1664. Camassa, R., Holm, D.D., and Hyman, J.M. (1994). A new integrable shallow water equation, Adv. Appl. Mech. 31, 1–33. Campbell, G.A. and Foster, R.M. (1948). Fourier Integrals for Practical Applications, Van Nostrand, New York. Canosa, J. (1969). Diffusion in nonlinear multiplicative media, J. Math. Phys. 10, 1862–1868. Canosa, J. (1973). On a nonlinear diffusion equation describing population growth, IBM J. Res. Dev. 17, 307–313.
818
Bibliography
Carpinteri, A. and Mainardi, F. (1997). Fractals and Fractional Calculus in Continuum Mechanics, Springer, New York. Case, K.M. and Chiu, C.S. (1969). Burgers’ turbulence models, Phys. Fluids 12, 1799–1808. Cercignani, C. (1977). Solitons. Theory and application, Rev. Nuovo Cimento 7, 429– 469. Chakraborty, S. and Kodama, Y. (2009). Soliton solutions of the KP equation and application to shallow water waves, Stud. Appl. Math. 123, 83–151. Champneys, A.R., Vanden-Broeck, J.M., and Lord, G.J. (2002). Do true elevation gravity-capillary solitary waves exist? A numerical investigation, J. Fluid Mech. 454, 403–417. Chen, Y. and Liu, P.L.-F. (1994). A pseudospectral approach for scattering of water waves, Proc. R. Soc. Lond. A445, 619–636. Chen, Y. and Liu, P.L.-F. (1995a). The unified Kadomtsev–Petviashvili equation for interfacial waves, J. Fluid Mech. 288, 383–408. Chen, Y. and Liu, P.L.-F. (1995b). Modified Boussinesq equations and associated parabolic models for water wave propagation, J. Fluid Mech. 228, 351–381. Chen, Y. and Wen, S. (1987). Traveling wave solutions to the two-dimensional Korteweg–de Vries equation, J. Math. Anal. Appl. 127, 226–236. Choi, W. and Camassa, R. (1999). Fully nonlinear internal waves in a two-fluid system, J. Fluid Mech. 396, 1–36. Choudhury, S.R. and Debnath, L. (1994). Painlevé analysis, similarity reductions, and special solutions of nonlinear evolution equations, in Ocean Waves Engineering (ed. M. Rahman), Vol. 2, 115–133. Christov, O. and Hakkaev, S. (2009). On the Cauchy problem for the periodic bfamily of equations and of the non-uniform continuity of Degasperis–Procesi equation, J. Math. Anal. Appl. 360, 47–56. Chu, V.H. and Mei, C.C. (1970). On slowly varying Stokes waves, J. Fluid Mech. 41, 873–887. Chu, V.H. and Mei, C.C. (1971). The nonlinear evolution of Stokes waves in deep water, J. Fluid Mech. 47, 337–351. Claude, Ch., Kivshar, Yu.S., Kluth, O., and Spataschek, K.H. (1993). Liapunov Stability of Generalized Langmuir Solitons, Phys. Rev. B47, 228–234. Coclite, G.M. and Karlsen, K.H. (2006). On the well-posedness of the Degasperis– Procesi equation, J. Funct. Anal. 233, 60–91. Cole, J.D. (1951). On a quasilinear parabolic equation occurring in aerodynamics, Q. Appl. Math. 9, 225–236. Collins, M.A. (1981). A quasi-continuum approach for solitons in an atomic chain, Chem. Phys. Lett. 77, 342–347. Constantin, A. (2000). Existence of permanent and breaking waves for a shallow water equation: A geometric approach, Ann. Inst. Fourier 50, 321–362. Constantin, A. (2001). On the scattering problem for the Camassa–Holm equation, Proc. R. Soc. Lond. A 457, 953–970. Constantin, A. and Escher, J. (1998a). Wave breaking for nonlinear nonlocal shallow water equations, Acta. Math. 181, 229–243.
Bibliography
819
Constantin, A. and Escher, J. (1998b). Well-posedness, global existence, and blowup phenomena for a periodic quasi-linear hyperbolic equation, Commun. Pure Appl. Math. 51, 475–504. Constantin, A. and Escher, J. (1998c). On the structure of family of quasilinear equations arising in shallow water theory, Math. Ann. 312, 403–416. Constantin, A. and Escher, J. (2000). On the blow-up rate and blow-up set of breaking waves for a shallow water equation, Math. Z. 233, 75–91. Constantin, A. and Lannes, D. (2009). The hydrodynamical relevance of the Camassa–Holm and Degaperis–Procesi Equation, Arch. Ration. Mech. Anal. 192, 165–186. Constantin, A. and McKean, H.P. (1999). A shallow water equation on the circle, Commun. Pure Appl. Math. 52(8), 949–982. Constantin, A. and Strauss, W. (2000). Stability of peakons, Commun. Pure Appl. Math. 53, 603–610. Constantin, A. and Strauss, W. (2002). Stability of the Camassa–Holm solitons, J. Nonlinear Sci. 12, 415–422. Constantin, A., Gerdjikov, V.S., and Ivanov, R.I. (2006). Inverse scattering transform for the Camassa–Holm equation, Inverse Probl. 22, 2197–2207. Cooper, F., Hyman, J.M., and Khare, A. (2001). Compacton solutions in a class of generalized fifth-order Korteweg–de Vries equations, Phys. Rev. E64, 1–13. Cornish, V. (1934). Ocean Waves and Kindred Geophysical Phenomena, Cambridge University Press, Cambridge. Courant, R. and Friedrichs, K.O. (1948). Supersonic Flow and Shock Waves, Interscience, New York and London Courant, R. and Hilbert, D. (1953). Methods of Mathematical Physics, Vol. 1, WileyInterscience, New York. Courant, R. and Hilbert, D. (1962). Methods of Mathematical Physics, Vol. 2, WileyInterscience, New York. Craig, W., Guyenne, P., and Kalisch, H. (2004). A new model for large amplitude long internal waves, C. R., Méc. 332, 525–530. Craik, A.D.D. (2004). The origins of water wave theory, Annu. Rev. Fluid Mech. 36, 1–28. Cramer, M.S. and Sen, R. (1992). A general scheme for the derivation of evolution equations describing mixed nonlinearity, Wave Motion 15, 333–355. Crandall, R.E. (1991). Mathematica for the Sciences, Addison-Wesley, Redwood City. Crank, J. (1975). The Mathematics of Diffusion, 2nd edition, Oxford University Press, Oxford. Crighton, D.G. (1979). Model equations of nonlinear acoustics, Annu. Rev. Fluid Mech. 11, 11–23. Crighton, D.G. and Scott, J.F. (1979). Asymptotic solution of model equations in nonlinear acoustics, Philos. Trans. R. Soc. Lond. A292, 101–134. Crowdy, D.G. (1999). Exact solutions for steady capillary waves on a fluid annulus, J. Nonlinear Sci. 9, 615–640. Das, A. (1989). Integrable Models, World Scientific, Singapore.
820
Bibliography
Davey, A. (1972). The propagation of a weak nonlinear wave, J. Fluid Mech. 53, 769–781. Davey, A. and Stewartson, K. (1974). On three dimensional packets of surface waves, Proc. R. Soc. Lond. A338, 101–110. Davis, H.T. (1962). Introduction to Nonlinear Differential and Integral Equations, Dover, New York. Davis, R.E. and Acrivos, A. (1967). Solitary internal waves in deep water, J. Fluid Mech. 29, 593–607. De Leonardis, R.M. and Trullinger, S.E. (1980). Theory of boundary effects on sineGordon solitons, J. Appl. Phys. 51, 1211–1226. Debnath, L. (1983a). Nonlinear Waves, Cambridge University Press, Cambridge. Debnath, L. (1983b). Nonlinear dynamics of water waves and the nonlinear Schrödinger equation, Diamond Jubilee Volume, Calcutta Mathematical Society, 63–93. Debnath, L. (1984). Advance in Nonlinear Waves, Vol. I, Pitman, London. Debnath, L. (1985a). Advance in Nonlinear Waves, Vol. II, Pitman, London. Debnath, L. (1985b). Nonlinear dispersive waves and variational principles, in Differential Geometry, Calculus of Variations and Their Applications (eds. G.M. Rassias and T.M. Rassias), Dekker, New York, 149–176. Debnath, L. (1987). Bifurcation and nonlinear instability in applied mathematics, in Nonlinear Analysis (ed. T.M. Rassias), World Scientific, Singapore, 161–266. Debnath, L. (1989). The linear and nonlinear Cauchy–Poisson wave problems for an inviscid or viscous liquid, in Topics in Mathematical Analysis (ed. T.M. Rassias), World Scientific, Singapore, 123–155. Debnath, L. (1992a). Nonlinear Dispersive Wave Systems, World Scientific, Singapore. Debnath, L. (1992b). Solitons and the inverse scattering transforms, Bull. Calcutta Math. Soc. 84, 475–504. Debnath, L. (1994). Nonlinear Water Waves, Academic Press, Boston. Debnath, L. (1995). Integral Transforms and Their Applications, CRC Press, Boca Raton. Debnath, L. (2000). Some nonlinear evolution equations in water waves, J. Math. Anal. Appl. 251, 488–503. Debnath, L. (2002). Nonlinear dynamics of water waves and breaking phenomena, Proc. Indian Natl. Sci. Acad., 683–753. Debnath, L. (2003a). Fractional integral and fractional partial differential equation in fluid mechanics, Fract. Calc. Appl. Anal. 6, 119–155. Debnath, L. (2003b). Recent applications of fractional calculus to science and engineering, Int. J. Math. Sci. 54, 3413–3442. Debnath, L. (2007a). A brief historical introduction to solitons and inverse scattering transform—a vision of Scott Russell, Int. J. Math. Educ. Sci. Technol. 38, 1003– 1028. Debnath, L. (2007b). On linear and nonlinear Rossby Waves, J. Math. Anal. Appl. 333, 164–190.
Bibliography
821
Debnath, L. (2009). Water waves and the Korteweg and de Vries equation, in Encyclopedia of Complexity and System Science, Vol. 23, Springer, Berlin, 9967– 10005. Debnath, L. and Bhatta, D. (2004). On solutions to a few linear fractional inhomogeneous partial differential equations in fluid mechanics, Fract. Calc. Appl. Anal. 7, 21–36. Debnath, L. and Bhatta, D. (2007). Integral Transforms and Their Applications, 2nd edition, Chapman & Hall/CRC Press, Boca Raton. Debnath, L. and Bhatti, M. (2004). On Fractional Schrödinger and Dirac equations, Int. J. Pure Appl. Math. 15, 1–11. Debnath, L. and Choudhury, S.R. (1997). Nonlinear Instability Analysis, Vol. 1, Computational Mechanics Publications, Southampton. Debnath, L. and Mikusinski, P. (1999). Introduction to Hilbert Spaces with Applications, 2nd edition, Academic Press, Boston. Debnath, L. and Mikusinski, P. (2005). Introduction to Hilbert Spaces with Applications, 3rd edition, Elsevier Academic Press, London. Debnath, L. and Rahman, M. (1981). A theory of nonlinear wave loading on offshore structures, Int. J. Math. Math. Sci. 4, 589–613. Debnath, L. and Shivamoggi, B.K. (1985). Some nonlinear model equations in dispersive systems and their solitary wave solutions in fluids and plasmas, in Advances in Nonlinear Waves II (ed. L. Debnath), Pitman Advance Publishing Program, Boston, 309–347. Debnath, L. and Shivamoggi, B.K. (1986). Three-dimensional nonlinear Schrödinger equation for finite-amplitude gravity waves in a fluid, Nuovo Cimento 94B, 140– 148. (1987) 99B, 247. Debnath, L. and Shivamoggi, B.K. (1990). Pseudo-variational principles and nonlinear surface waves in a dissipative fluid, Int. J. Non-Linear Mech. 25, 61–65. Defermos, C.M. (2000). Hyperbolic Conservation Laws in Continuum Physics, Springer, Berlin. Degasperis, A. (2010). Integrable models in nonlinear optics and soliton solutions, J. Phys. A 43(43), 1–18, #434001. Degasperis, A. and Procesi, M. (1999). Asymptotic Integrability, in Symmetry and Perturbation Theory (eds. A. Degasperis and G. Gaeta), World Scientific, Singapore. Degasperis, A., Holm, D., and Hone, A. (2002). A new integral equation with peakon solutions, Theor. Math. Phys. 133, 1461–1472. Degasperis, A., Conforti, M., Baronio, F., and Wabnitz, S. (2007). Effects of nonlinear wave coupling: accelerate solitons, Eur. Phys. J. Spec. Top. 147, 233–252. Dias, F. and Bridges, T. (2005). Weakly nonlinear wave packets and the nonlinear Schrödinger equation (Chapter 2), in Nonlinear Waves in Fluids: Recent Advances and Modern Applications (ed. R. Grimshaw), CISM Courses and Lectures, Vol. 483, Springer, Wien and New York, 29–67. Dias, F. and Kharif, C. (1999). Nonlinear gravity and capillary-gravity waves, Annu. Rev. Fluid Mech. 31, 301–346.
822
Bibliography
Dias, F. and Milewski, P. (2010). On the fully-nonlinear shallow-water generalized Serre equations, Phys. Lett. A 374, 1049–1053. Dickey, L.A. (2000). Soliton Equations and Hamiltonian Systems, World Scientist, Singapore. Dinda Tchofo, P. and Remoissenet, M. (1999). Breather compactons in nonlinear Klein–Gordon systems, Phys. Rev. E60, 6218–6221. Ditkin, V.A. and Prudnikov, A.P. (1965). Integral Transforms and Operational Calculus, Pergamon Press, Oxford. Dodd, R.K. and Fordy, A.P. (1983). The prolongation structures of quasipolynomial flows, Proc. R. Soc. Lond. A385, 389–429. Dodd, R.K. and Fordy, A.P. (1984). Prolongation structures of complex quasipolynomial evolution equations, J. Phys. A, Math. Gen. 17, 3249–3266. Dodd, R.K. and Gibbon, J.D. (1997). The prolongation structures of a higher order Korteweg–de Vries equation, Proc. R. Soc. Lond. A358, 287–296. Doetsch, G. (1970). Introduction to the Theory and Applications of the Laplace Transformation, Springer, New York. Drazin, P.G. (1983). Solitons, Cambridge University Press, Cambridge. Drazin, P.G. and Johnson, R.S. (1989). Solitons: An Introduction, Cambridge University Press, Cambridge. Dresner, L. (1983). Similarity Solutions of Nonlinear Partial Differential Equations, Pitman Books, London Dressler, R.F. (1949). Mathematical solution of problem of roll waves in inclined open channels, Commun. Pure Appl. Math. 2, 149–194. Dullin, H., Gottwald, G., and Holm, D. (2003). Camassa–Holm, Korteweg–de Vries5 and other asymptotically equivalent shallow water waves, Fluid Dyn. Res. 33, 73–95. Dullin, H., Gottwald, G., and Holm, D. (2004). On asymptotically equivalent shallow water wave equations, Physica D 190, 1–14. Dunbar, S.R. (1983). Travelling wave solutions of diffusive Lotka–Volterra equations, J. Math. Biol. 17, 11–32. Duncan, D.B., Eilbeck, J.C., Feddersen, H., and Wattis, J.A.D. (1993). Solitons on Lattices, Physica D 68, 1–11. Dusuel, S., Michaux, P., and Remoissnet, M. (1998). From kinks to compactionlike kinks, Phys. Rev. E57, 2320–2326. Dutta, M. and Debnath, L. (1965). Elements of the Theory of Elliptic and Associated Functions with Applications, World Press Pub. Ltd., Calcutta. Eilbeck, J.C. and Flesch, B. (1990). Calculation of families of solitary waves on discrete lattices, Phys. Lett. A149, 200–202. El, G.A. (2007). Korteweg–de Vries equation: Solitons and undular bores, in Solitary Waves in Fluids (ed. R.J.H. Grimshaw), 19–54. El, G.A., Grimshaw, R.H.J., and Pavlov, M.V. (2001). Integrable shallow water equations and undular bores, Stud. Appl. Math. 107, 157–186. El-Sherbiny, H.M. and Debnath, L. (1997). Equivalence between Bäcklund transformation and inverse scattering methods for a class of nonlinear differential equations, Int. J. Non-Linear Mech. 31, 51–58.
Bibliography
823
Engelbrecht, J. (1981). On theory of pulse transmission in a nerve fibre, Proc. R. Soc. Lond. A375, 195–209. Engelbrecht, J. and Peipman, T. (1992). Nonlinear waves in a layer with energy influx, Wave Motion 16, 173–181. Engelbrecht, J.K., Fridman, V.E., and Pelinovski, E.N. (1988). Nonlinear Evolution Equations, Longman, London. Enz, U. (1963). Discrete mass, elementary length, and a topological invariant as a consequence of relativistic invariant principles, Phys. Rev. 131, 1392–1394. Erdélyi, A. (1955). Higher Transcendental Functions, Vol. 3, McGraw-Hill, New York. Erdélyi, A., Magnus, W., Oberhettinger, F., and Tricomi, F. (1954). Tables of Integral Transforms, Vols. 1 and 2, McGraw-Hill, New York. Escher, J., Liu, Y., and Yin, Z. (2006). Global weak solution and blow-up structure for the Degasperis–Procesi equation, J. Funct. Anal. 241, 457–485. Estrada, R. and Kanwal, R.P. (1994). Asymptotic Analysis, Birkhäuser, Boston. Evans, D.J. and Sahimi, M.S. (1989). The alternative group explicit (AGE) iterative method to solve parabolic and hyperbolic partial differential equations, Annu. Numer. Fluid Mech. Heat Transf. 2, 259–283. Fang, G.C., Chen, W.Y., and Hu, Z. (2008). Well-posedness of the solution to the D-P equation with dispersive term, Int. J. Nonlinear Sci. 5, 125–132. Feigenbaum, M.J. (1978). Qualitative universality for a class of nonlinear transformation, J. Stat. Phys. 19, 25–52. Fermi, E., Pasta, J., and Ulam, S. (1955). Studies of nonlinear problems I, Los Alamos Report LA 1940, in Lectures in Applied Mathematics (ed. A.C. Newell), Vol. 15, Am. Math. Soc., Providence, 143–156. Fickett, W. and Davies, W.C. (1979). Detonation, University of California Press, Berkeley. Fife, P.C. (1979). Mathematical Aspects of Reacting and Diffusing Systems, Lecture Notes in Biomathematics, Vol. 28, Springer, Berlin. Fisher, R.A. (1936). The wave of advance of advantageous genes, Ann. Eugen. 7, 335–369. FitzHugh, R. (1961). Impulses and physiological states in theoretical models of nerve membrane, Biophys. J. 1, 445–466. Flaschka, H. (1974a). The Toda lattice: II. Existence of Integrals, Phys. Rev. B 9, 1924–1925. Flaschka, H. (1974b). On the Toda lattice: II. Inverse-scattering solutions, Prog. Theor. Phys. 51, 703–716. Fokas, A.S. (2008). A Unified Approach to Boundary Value Problems, SIAM, Philadelphia. Fokas, A.S. and Ablowitz, M.J. (1981). Linearization of the Korteweg–de Vries and Painlevé II equations, Phys. Rev. Lett. 47, 1096–1100. Fokas, A.S. and Ablowitz, M.J. (1983). The inverse scattering transform for the Benjamin–Ono equation-pivot to multidimensional equations, Stud. Appl. Math. 68, 1–10. Fokas, A.S. and Fuchssteiner, B. (1981). Symplectic structures, their bäcklund transformations and hereditary symmetries, Physica D 4, 47–66.
824
Bibliography
Ford, J., Stoddard, S.D., and Turner, J.S. (1973). On the integrability of the Toda lattice, Prog. Theor. Phys. 50, 1547–1560. Fordy, A.P. (1990). Prolongation structures of nonlinear evolution equations, in Soliton Theory: A Survey of Results (ed. A.P. Fordy), Manchester University Press, Manchester, 403–404. Fornberg, B. and Whitham, G.B. (1978). A numerical and theoretical study of certain nonlinear wave phenomena, Philos. Trans. R. Soc. Lond. A289, 373–404. Freeman, N.C. (1980). Soliton interactions in two dimensions, Adv. Appl. Mech. 20, 1–37. Freeman, N.C. (1984). Soliton solutions of nonlinear evolution equations, IMA J. Appl. Math. 32, 125–145. Freeman, N.C. and Nimmo, J.J.C. (1983). Soliton-solutions of the Korteweg–de Vries and Kadomtsev–Petviashvili equations: the Wronskian technique, Phys. Lett. A 95, 1–3. Frenkel, J. and Kontorova, T. (1939). On the theory of plastic deformation and twinning, Fix. Zh. 1, 137–139. Frieman, E.A. (1963). On a new method in the theory of irreversible process, J. Math. Phys. 4, 410–418. Fuchssteiner, B. (1981). The Lie algebra structure of nonlinear evolution equations and infinite dimensional Abelian symmetry groups, Prog. Theor. Phys. 65, 861– 876. Fulton, T.A. (1972). Aspects of vortices in long Josephson junctions, Bull. Am. Phys. Soc. 17, 46–49. Fusco, D. and Manganaro, N. (1994). A method for determining exact solutions to a class of nonlinear models based on introduction of differential constraints, J. Math. Phys. 35, 3659–3667. Fusco, D. and Manganaro, N. (1996). A method for finding exact solutions to hyperbolic systems of first-order partial differential equations, IMA J. Appl. Math. 57, 223–240. Garabedian, P.R. (1986). Partial Differential Equations, 2nd edition, Chelsea, New York. Gardner, C.S. and Morikawa, G.K. (1960). Similarity in the asymptotic behavior of collision-free hydromagnetic waves and water waves, Courant Inst. Math. Sci. Rep. NYO-9082, 1–30. Gardner, C.S., Greene, J.M., Kruskal, M.D., and Miura, R.M. (1967). Method for solving the KdV equation, Phys. Rev. Lett. 19, 1095–1097. Gardner, C.S., Greene, J.M., Kruskal, M.D., and Miura, R.M. (1974). Korteweg– de Vries equation and generalizations, VI, Methods for exact solution, Commun. Pure Appl. Math. 27, 97–133. Gazdag, J. and Canosa, J. (1974). Numerical solution of Fisher’s equation, J. Appl. Probab. 11, 445–457. Gazis, D.C. (1974). Traffic Science, John Wiley, New York. Gelfand, I.M. and Levitan, B.M. (1951). On the determination of a differential equation from its spectral function, Am. Math. Transl. (2) 1, 253–304.
Bibliography
825
Gentile, G. and Procesi, M. (2009). Periodic solutions for a class of nonlinear partial differential equations in higher dimension, Commun. Math. Phys. 289, 863–906. Ghez, R. (1988). A Primer of Diffusion Problems, Wiley, New York. Ghosh, K.K. and Debnath, L. (1997). Some exact solutions of nonlinear shallow water equations, Int. J. Nonlinear Mech. 31, 104–108. Gibbon, J.D. (1985). A survey of the origins and physical importance of soliton equations, Philos. Trans. R. Soc. Lond. A315, 335–365. Gibbon, J.D., James, I.N. and Moroz, I.M. (1979). An example of soliton behaviour in a rotating fluid, Proc. R. Soc. Lond. A367, 185–219. Gibbons, J., Thornhill, S.G., Wardrop, M.J., and ter Haar, D. (1977). On the theory of Langmuir solitons, J. Plasma Phys. 17, 153–170. Gillbarg, D. and Trudinger, N.S. (1983). Elliptic Partial Differential Equations of Second Order, 2nd edition, Springer, Berlin. Goldstein, S. (1953). On the mathematics of exchange processes in fixed columns, Parts I and II, Proc. R. Soc. Lond. A219, 151–185. Goldstein, H. (1965). Classical Mechanics, Addison-Wesley, Cambridge. Goldstein, S. and Murray, J.D. (1959). On the mathematics of exchange processes in fixed colums, Parts III, IV, V, Proc. R. Soc. Lond. A252, 334–375. Gómez-Ullate, et al. (eds.) (2010). Current trends in integrability and nonlinear phenomena, J. Phys. A. Math. Theor. 43, IOP Publishing, Bristol. Gordon, W. (1926). Der Comptoneffekt nach der Schrödingerchen Theorie, Zeit. Für Physik 40, 117–133. Gorschkov, K.A., Ostrovsky, L.A., and Pelinovsky, E.N. (1974). Some problems of asymptotic theory of nonlinear waves, Proc. IEEE 62, 1511–1517. Greenberg, H. (1959). An analysis of traffic flow, Oper. Res. 7, 79–85. Grimshaw, R. (1985). Evolution equations for weakly nonlinear, long internal waves in a rotating fluid, Stud. Appl. Math. 73, 1–33. Grimshaw, R. (2007a). Solitary Waves in Fluids, WIT Press, Southampton. Grimshaw, R.H.J. (2007b). Envelope solitary waves, in Solitary Waves in Fluids (ed. R.H.J. Grimshaw), 159–174. Grimshaw, R., Pelinovsky, D., Pelinovsky, E. and Slunyaev, A. (2002). Generation of large-amplitude solitons in the extended Korteweg–de Vries equation, Chaos 12, 1070–1076. Grindrod, P. (1991). Patterns and Waves, Oxford University Press, Oxford. Groves, M.D. (2004). Steady water waves, J. Nonlinear Math. Phys. 11, 435–460. Gui, G., Lui, Y., and Tian, L. (2008). Global existence and blow-up phenomena for the peakon b-family of equations, Indiana Univ. Math. J. 57, 1209–1234. Guo, F. (2009). Global weak solutions and wave breaking phenomena to the periodic Degasperis–Procesi equation with strong dispersion, Nonlinear Anal. 71, 5280– 5295. Gao, F. (2009). Global weak solutions and breaking waves to the Degasperis–Procesi equation with linear dispersion, J. Math. Anal. Appl. 360, 345–362. Gurney, W.S.C. and Nisbet, R.M. (1975). The regulation of inhomogeneous populations, J. Theor. Biol. 52, 441–457. Gurtin, M.E. and MacCamy, R.C. (1977). On the diffusion of biological populations, Math. Biosci. 33, 35–49.
826
Bibliography
Haberman, R. (1977). Mathematical Models: Mechanical Vibrations, Population Dynamics and Traffic Flow, Prentice-Hall, Englewood Cliffs. Haberman, R. (1988). The modulated phase shift for weakly dissipated nonlinear oscillatory waves of the Korteweg–de Vries type, Stud. Appl. Math. 78, 73–90. Hagstrom, T. and Keller, H.B. (1986). The numerical calculations of traveling wave solutions of nonlinear parabolic equations (Preprint) Haight, F.A. (1963). Mathematical Theories of Traffic Flow, Academic Press, New York. Hammack, J.L. and Segur, H. (1974). The Korteweg–de Vries equation and water waves, Part 2, Comparison with experiments, J. Fluid Mech. 65, 289–314. Hansen, A.G. (1964). Similarity Analyses of Boundary Value Problems in Engineering, Prentice-Hall, Englewood Cliffs. Hasegawa, A. (1990). Optical Solitons in Fibers, 2nd edition, Springer, Berlin. Hasegawa, A. and Tappert, F. (1973). Transmission of stationary nonlinear optical pulses in dispersive dielectric fibers I: Anomalous dispersion, Appl. Phys. Lett. 23, 142–144. Hasimoto, H. (1972). A soliton on a vortex filament, J. Fluid Mech. 51, 477–485. Hasimoto, H. and Ono, H. (1972). Nonlinear modulation of gravity waves, J. Phys. Soc. Jpn. 33, 805–811. Hayes, W.D. (1973). Group velocity and nonlinear dispersive wave propagation, Proc. R. Soc. Lond. A332, 199–221. Henon, M. (1974). Integrals of the Toda lattice, Phys. Rev. B 9, 1921–1923. Henry, D. (2005). Infinite propagation speed for the Degasperis–Procesi equation, J. Math. Anal. Appl. 311, 755–759. Herman, R.L. (1990). Resolution of the motion of a perturbed KdV solition, Inverse Probl. 6, 43–54. Herman, R., Montroll, E.W., Potts, R.B., and Rothery, R.W. (1959). Traffic dynamics: Analysis of stability in a car following, Oper. Res. 7, 86–106. Hernandez, J. (1986). Qualitative methods for nonlinear differential equations, in Nonlinear Diffusion Problems, Lecture Notes in Mathematics, Vol. 1224, Springer, New York, 47–118. Himonas, A. and Misiolek, G. (2005). High-frequency smooth solutions and wellposedness of the Camassa–Holm equation, Int. Math. Res. Not. IMRN 51, 3135– 3151. Hiraoka, Y. and Kodama, Y. (2009). Normal forms and solitons, in Integrability (ed. A.V. Mikhailov), Springer, New York. Hirota, R. (1971). Exact solution of the Korteweg–de Vries equation for multiple collisions of solitons, Phys. Rev. Lett. 27, 1192–1194. Hirota, R. (1973a). Exact envelope-soliton solutions of a nonlinear wave equation, J. Math. Phys. 14, 805–809. Hirota, R. (1973b). Exact N -solutions of the wave equation of long waves in shallow water and in nonlinear lattices, J. Math. Phys. 14, 810–814. Hirota, R. (1981). Discrete analogue of a generalized Toda equation, J. Phys. Soc. Jpn. 50, 3785–3791. Hirota, R. (2004). The Direct Method in Soliton Theory, Cambridge University Press, Cambridge.
Bibliography
827
Hirota, R. and Suzuki, K. (1970). Studies on lattice solitons by using electrical networks, J. Phys. Soc. Jpn. 28, 1366–1367. Hirota, R. and Suzuki, K. (1973). Theoretical and experimental studies of lattice solitons in nonlinear lumped networks, Proc. IEEE 61, 1483–1491. Hirota, R., Ito, M., and Kako, F. (1988). Two-dimensional Toda Lattice Equations, Prog. Theor. Phys. Suppl. 94, 42–58. Hodgkin, A.L. and Huxley, A.F. (1952). A qualitative description of membrane current and its applications to conduction and excitation in nerve, J. Physiol. (Lond.) 117, 500–544. Holden, H. and Risebro, N.H. (2002). Front Tracking for Hyperbolic Conservation Laws, Springer, New York. Holliday, D. (1973). Nonlinear gravity-capillary surface waves in a slowly varying current, J. Fluid Mech. 57, 797–802. Holm, D.D. and Ivanov, R.I. (2010). Smooth and peaked solitons of the CH equation, J. Phys. A, Math. Theor. 43(43), 1–18, #434003. Holm, D.D. and Staley, M.F. (2003). Nonlinear balance and exchange of stability in dynamics of solitons, peakons, ramps/cliffs and leftons in a 1 + 1 nonlinear evolutionary PDE, Phys. Lett. A 308, 437–444. Hopf, E. (1950). The partial differential equation ut + uux = μuxx , Commun. Pure Appl. Math. 3, 201–230. Hoppensteadt, F.C. (1975). Mathematical Theories of Populations: Demographics, Genetics, and Epidemic, CBMS Lectures, Vol. 20, SIAM, Philadelphia. Hunter, J.K. and Scherule, J. (1988). Existence of perturbed solitary wave solutions to a model equation for water waves, Physica D 32, 253–268. Hunter, J.K. and Vanden-Broeck, J.M. (1983). Solitary and periodic gravity-capillary waves of finite amplitude, J. Fluid Mech. 134, 205–219. Ichikawa, Y.H. (1979). Topics on solitons in plasmas, Phys. Scr. 20, 296–305. Ichikawa, V.H., Imamura, T., and Taniuti, T. (1972). Nonlinear wave modulation in collisionless plasma, J. Phys. Soc. Jpn. 33, 189–197. Iizuka, T. and Wadati, M. (1992). Shallow water waves over an uneven bottom and an inhomogeneous KP equation, Chaos Solitons Fractals 2, 575–582. Ilichev, A.T. and Marchenko, A.B. (1989). Propagation of the long nonlinear waves in a ponderable fluid beneath an ice sheet, Fluid Dyn. 24, 88–95. Infeld, E. (1980). On three-dimensional generalizations of the Boussinesq and Korteweg–de Vries equations, Q. Appl. Math. XXXVIII, 277–287. Infeld, E. and Rowlands, G. (1990). Nonlinear Waves, Solitons and Chaos, Cambridge University Press, Cambridge. Infeld, E., Ziemkiewicz, J., and Rowlands, G. (1987). The two surface dimensional stability of periodic and solitary wavetrains on a water surface over arbitrary depth, Phys. Fluids 30, 2330–2338. Inoue, Y. and Matsumoto, Y. (1974). Nonlinear wave modulation in dispersive media, J. Phys. Soc. Jpn. 36, 1446–1455. Iooss, G. and Kirchgassner, K. (1990a). Bifurcation d’ondes solitaires en pre´sence d’une faible tension superficielle, C. R. Acad. Sci., Paris 311, 265–268. Iooss, G. and Kirchgassner, K. (1990b). Water waves for small surface tension: an approach via normal form, Proc. R. Soc. Edinb. 122A, 267–299.
828
Bibliography
Jeffrey, A. (1976). Quasilinear Hyperbolic Systems and Waves, Research Notes in Mathematics, Vol. 5, Pitman Publishing Company, London. Jeffrey, A. and Engelbrecht, J. (1994). Nonlinear Waves in Solids, Springer, New York. Jeffrey, A. and Kakutani, T. (1972). Weak nonlinear dispersive waves: A discussion centered around the KdV equation, SIAM Rev. 14, 582–643. Jeffrey, A. and Kawahara, T. (1982). Asymptotic Methods in Nonlinear Wave Theory, Pitman Advance Publishing Program, Boston. Jeffrey, A. and Taniuti, T. (1964). Nonlinear Wave Propagation, Academic Press, New York. Jeffrey, A. and Xu, S. (1989). Exact solution to the Korteweg–de Vries–Burgers equation, Wave Motion 11, 559–564. John, F. (1982). Partial Differential Equations, 4th edition, Springer, New York. Johnson, R.S. (1970). A nonlinear equation incorporating damping and dispersion, J. Fluid Mech. 42, 49–60. Johnson, R.S. (1980). Water waves and Korteweg–de Vries equations, J. Fluid Mech. 97, 701–719. Johnson, R.S. (1983). The KdV equation and related problems in water wave theory, in Nonlinear Waves (ed. L. Debnath), Cambridge University Press, Cambridge, 25–43. Johnson, R.S. (1997). A Modern Introduction to the Mathematical Theory of Water Waves, Cambridge University Press, Cambridge. Johnson, R. (2002). Camassa–Holm, Korteweg–de Vries and related models for water waves, J. Fluid Mech. 455, 63–82. Johnson, R. (2003). On solutions of the Camassa–Holm Equation, in Proceedings: Mathematical, Physical and Engineering Sciences, Vol. 459 (No. 2035). Jones, D.S. (1966). Generalized Functions, Academic Press, New York. Jones, D.S. (1982). Theory of Generalized Functions, Cambridge University Press, Cambridge. Jones, D.S. and Sleeman, B.D. (1983). Differential Equations and Mathematical Biology, Allen and Unwin, London. Josephson, B.D. (1965). Supercurrents through the barriers, Adv. Phys. 14, 419–451. Kadomtsev, B.B. and Petviashvili, V.I. (1970). On stability of solitary waves in weakly dispersive media, Dokl. Akad. Nauk SSSR 192, 753–756; Sov. Phys. Dokl. 15, 539–541. Kako, F. and Yajima, N. (1980). Interaction of ion-acoustic solitons in multidimensional space, J. Phys. Soc. Jpn. 49, 2063–2071. Kakutani, T. (1971a). Effect of an uneven bottom on gravity waves, J. Phys. Soc. Jpn. 30, 272–276. Kakutani, T. (1971b). Weak nonlinear magneto-acoustic waves in an inhomogeneous plasma, J. Phys. Soc. Jpn. 31, 1246–1248. Kakutani, T. and Ono, H. (1969). Weak nonlinear hydromagnetic waves in a cold collision-free plasma, J. Phys. Soc. Jpn. 26, 1305–1318. Kakutani, T., Ono, H., Taniuti, T., and Wei, C.C. (1968). Reductive perturbation method in nonlinear wave propagation II. Application to hydromagnetic waves in cold plasma, J. Phys. Soc. Jpn. 24, 1159–1166.
Bibliography
829
Kaliappan, P. (1984). An exact solution for travelling waves, Physica D11, 368–374. Kambe, T. (2007). Elementary Fluid Mechanics, World Scientific, Singapore. Kametaka, Y. (1976). On the nonlinear diffusion equation of Kolmogorov– Petrovskii–Piskunov type, Osaka J. Math. 13, 11–66. Karpman, V.I. (1967). An asymptotic solution of the Korteweg–de Vries equations, Phys. Lett. 25A, 708–709. Karpman, V.I. (1971). High frequency electromagnetic field in plasma with negative dielectric constant, Plasma Phys. 13, 477–490. Karpman, V.I. (1975a). Nonlinear Waves in Dispersive Media, Pergamon Press, London. Karpman, V.I. (1975b). On the dynamics of sonic-Langmuir solitons, Phys. Scr. 11, 263–265. Karpman, V.I. and Krushkal, E.M. (1969). Modulated waves in nonlinear dispersive media, Sov. Phys. JETP 28, 277–281. Karpman, V.I. and Maslov, E.M. (1978). Structure of tails produced under the action of perturbations on solitons, Sov. Phys. JETP 48, 252–258. Kato, T. (1987). On nonlinear Schrödinger equations, Ann. Inst. Henri Poincaré, Phys. Théor. 46, 113–129. Katsis, C. and Akylas, T.R. (1987). Solitary internal waves in a rotating channel: A numerical study, Phys. Fluids 30, 297–301. Kaup, D.J. (1980). The Estabrook–Wahlquist method with examples of applications, Physica ID, 391–411. Kaup, D.J. and Newell, A.C. (1978). Solitons as particles and oscillators and in slowly changing media: A singular perturbation theory, Proc. R. Soc. Lond. A361, 413–446. Kaup, D.J. and Van Gorder, R.A. (2010). The inverse scattering transform and squared eigenfunctions for the nondegenerate 3 × 3 operator and its soliton structure, Inverse Probl. 26, 055005. Kawahara, T. (1972). Oscillatory solitary waves in dispersive media, J. Phys. Soc. Jpn. 33, 260–264. Kawahara, T. (1973). The derivative-expansion method and nonlinear dispersive waves, J. Phys. Soc. Jpn. 35, 1537–1544. Kawahara, T. (1975a). Nonlinear self-modulation of capillary-gravity waves on liquid layer, J. Phys. Soc. Jpn. 38, 265–270. Kawahara, T. (1975b). Derivative-expansion method for nonlinear waves on a liquid layer of slowly varying depth, J. Phys. Soc. Jpn. 38, 1200–1206. Kawahara, T. and Toha, S. (1987). Pulse interactions in an unstable dissipative dispersive nonlinear system, Phys. Fluids 31, 2103–2111. Keady, G. and Norbury, J. (1978). On the existence theory for irrotational water waves, Math. Proc. Camb. Philos. Soc. 83, 137–157. Kelley, P.L. (1965). Self-focusing of laser beams, Phys. Rev. Lett. 15, 1005–1008. Kevorkian, J. (1990). Partial Differential Equations, Brooks/Cole, Pacific Grove. Kevorkian, J.K. and Cole, J.D. (1996). Multiple Scale and Singular Perturbation Methods, Springer, New York. Kinnersley, W. (1976). Exact large amplitude capillary waves on sheets of fluid, J. Fluid Mech. 77, 229–241.
830
Bibliography
Kivshar, Y. (1993). Intrinsic localized modes as solitons with a compact support, Phys. Rev. B48, R43–R45. Kivshar, Y.S. and Malomed, B.A. (1989). Dynamics of solitons in nearly integrable systems, Rev. Mod. Phys. 61, 763–916. Klein, O. (1927). Elektrodynamik und Willenmechanik vom Standpunkt des Korrespondenzprinzips, Zeit. Für Physik 41, 407–442. Ko, K. and Kuehl, H.H. (1980). Energy loss of Korteweg–de Vries solitary wave in a slowly varying medium, Phys. Fluids 23, 834–836. Kodama, Y. (1985). Normal forms for weakly dispersive wave equations, Phys. Lett. A 112, 193–196. Kodama, Y. (2010). KP Solitons in shallow water, J. Phys. A, Math. Theor. 43(43), 1–54, #434004. Kolev, B. (2007). Bi-Hamiltonian systems on the dual of the Lie Algebra of vector fields of the circle and periodic shallow water equations, Philos. Trans. R. Soc. Lond. Ser. A 365, 2333–2357. Kolmogorov, A., Petrovsky, I., and Piscunov, N. (1937). A study of the equation of diffusion with increase in the quantity of matter and its application to a biological problem, Bull. Univ. Moscow, Ser. Int. Sec. A1, 1–25. Konno, K. and Wadati, M. (1975). Simple derivation of the Bäcklund transformation from the Riccati form of the inverse method, Prog. Theor. Phys. 53, 1652–1656. Kopell, N. and Howard, L.N. (1973). Plane wave solutions to reaction–diffusion equations, Stud. Appl. Math. 42, 291–328. Korteweg, D.J. and de Vries, G. (1895). On the change of form of long waves advancing in a rectangular canal, and on a new type of long stationary waves, Philos. Mag. (5) 39, 422–443. Krasovskii, Y.P. (1960). On the theory of steady waves of not all small amplitude (in Russian), Dokl. Akad. Nauk SSSR 130, 1237–1252. Krasovskii, Y.P. (1961). On the theory of permanent waves of finite amplitude, Zh. Vychisl. Mat. Fiz 1, 836–855 (in Russian). Kriess, H.O. and Lorenz, J. (1989). Initial-Boundary Value Problems and the Navier– Stokes Equations, Academic Press, New York. Kruskal, M. (1975). Nonlinear wave equations, in Dynamical Systems, Theory and Applications (ed. J. Moser), Lecture Notes in Physics, Vol. 38, Springer, Heidelberg, 310–354. Kulikovskii, A.G. and Slobodkina, F.A. (1967). Equilibrium of arbitrary steady flows at the transonic points, Prikl. Mat. Meh. 31, 623–630. Kundu, A. (1984). Landau–Lifschitz and higher order nonlinear systems gauge generated from nonlinear Schrödinger type equations, J. Math. Phys. 25, 3433–3438. Kundu, A. (1987). Exact solutions to higher order nonlinear equations through gauge transformations, Physica 25D, 399–406. Kuramoto, Y. (1984). Chemical Oscillations, Waves and Turbulence, Vol. 19, Springer, New York. Kynch, G.F. (1952). A theory of sedimentation, Trans. Faraday Soc. 48, 166–176. Lai, S. and Wu, Y. (2011). A model containing both the Camassa–Holm and Degasperis–Procesi equations, J. Math. Anal. Appl. 374, 458–469.
Bibliography
831
Lake, B.M., Yuen, H.C., Rundgaldier, H., and Ferguson, W.E. (1977). Nonlinear deep-water waves: Theory and experiment, Part 2, Evolution of a continuous wave train, J. Fluid Mech. 83, 49–74. Lakshmanan, M. (2007). Nonlinear wave equations and possible connections to tsunami dynamics, in Tsunami and Nonlinear Waves, Springer, New York. Lamb, H. (1904). On the propagation of tremors over the surface of elastic solid, Philos. Trans. R. Soc. Lond. A203, 1–42. Lamb, H. (1932). Hydrodynamics, 6th edition, Cambridge University Press, Cambridge. Lamb, G.L. (1967). Propagation of ultrashort optical pulses, Phys. Lett. 25A, 181– 182. Lamb, G.L. (1971). Analytical descriptions of ultrashort optical pulse propagation in a resonant medium, Rev. Mod. Phys. 49, 99–124. Lamb, G.L. (1973). Phase variation in coherent-optical-pulse propagation, Phys. Rev. Lett. 31, 196–199. Lamb, G.L. (1980). Elements of Soliton Theory, Wiley, New York. Landau, L.D. and Lifshitz, E.M. (1959). Fluid Mechanics, Pergamon Press, New York. Langerstrom, P.A., Cole, J.D., and Trilling, L., (1949). Problems in theory of visous compressible fluids, Calif. Inst. Tech. Monograph, Pasadena. Lardner, R.W. (1976). The development of shock waves in nonlinear viscoelastic media, Proc. R. Soc. Lond. A347, 329–344. Lardner, R.W. (1986). Third order solutions of Burgers equation, Q. Appl. Math. 44, 293–302. Lardner, R.W. and Arya, J.C. (1980). Two generalisations of Burgers equation, Acta Mech. 37, 179–190. Larson, D.A. (1978). Transient bounds and time asymptotic behaviour of solutions, SIAM J. Appl. Math. 34, 93–103. Lax, P.D. (1954a). Weak solutions of nonlinear hyperbolic equations and their numerical computation, Commun. Pure Appl. Math. 7, 159–193. Lax, P.D. (1954b). The initial-value problem for nonlinear hyperbolic equations in two independent variables, Ann. Math. Stud. (Princeton) 33, 211–299. Lax, P.D. (1957). Hyperbolic systems of conservation law, II, Commun. Pure Appl. Math. 10, 537–566. Lax, P.D. (1964). Development of singularities of solutions of nonlinear hyperbolic partial differential equations, J. Math. Phys. 5, 611–613. Lax, P.D. (1968). Integrals of nonlinear equations of evolution and solitary waves, Commun. Pure Appl. Math. 21, 467–490. Lax, P.D. (1972). The formation and decay of shock waves, Am. Math. Mon. 79, 227–241. Lax, P.D. (1973). Hyperbolic Systems of Conservation Laws and the Mathematical Theory of Shock Waves, Society for Industrial and Applied Mathematics, Philadelphia. Lax, P.D. (1975). Periodic solutions of the Korteweg–de Vries equation, Commun. Pure Appl. Math. 28, 141–188.
832
Bibliography
Lax, P.D. (1978). A Hamiltonian approach to the KdV and other equations, in Nonlinear Evolution Equations (ed. M. Crandall), Academic Press, New York, 207– 224. Lax, P.D. (2006). Hyperbolic Partial Differential Equations, Am. Math. Soc., Providence. Leibovich, S. (1970). Weakly nonlinear waves in rotating fluids, J. Fluids Mech. 42, 803–822. Leibovich, S. and Seebass, A.R. (1972). Nonlinear Waves, Cornell University Press, Ithaca. Lenells, J. (2005). Conservation laws of the Camassa–Holm equation, J. Phys. A 38, 869–880. Le Veque, R.J. (1990). Numerical Methods for Conservation Laws, Birkhäuser Verlag, Boston. Leo, M., Leo, R.A., Soliani, G., and Solombrino, L. (1983). Lie–Bäcklund symmetrics for the Harry Dym equation, Phys. Rev. D 27, 1406–1408. Li, Y.A. (2006). A shallow-water approximation to the full water wave problem, Commun. Pure Appl. Math. 59, 1225–1285. Li, Y.A. and Olver, P.J. (1997). Convergence of solitary-wave solutions in a perturbed bihamiltonian dynamical system: I. Compactons and peakons, Discrete Contin. Dyn. Syst. 3(3), 419–427. Li, Y.A. and Olver, P.J. (2000). Well-posedness and blow-up solutions for an integrable nonlinearly dispersive model wave equation, J. Differ. Equ. 162, 27–63. Lighthill, M.J. (1956). Viscosity effects in sound waves of finite amplitude, in Surveys in Mechanics (eds. G.K. Batchelor and R.M. Davies), Cambridge University Press, Cambridge, 250–351. Lighthill, M.J. (1957). River waves, Naval hydrodynamics publication 515, National Academy of Sciences–National Research Council, Washington, D.C. Lighthill, M.J. (1958). An Introduction to Fourier Analysis and Generalized Functions, Cambridge University Press, Cambridge. Lighthill, M.J. (1965). Group velocity, J. Inst. Math. Appl. 1, 1–28. Lighthill, M.J. (1967). Some special cases treated by the Whitham theory, Proc. R. Soc. Lond. A299, 38–53. Lighthill, M.J. (1978). Waves in Fluids, Cambridge University Press, Cambridge. Lighthill, M.J. and Whitham, G.B. (1955). On kinematic waves: I. Flood movement in long rivers; II. Theory of traffic flow on long crowded roads, Proc. R. Soc. Lond. A229, 281–345. Lin, Z.W. and Liu, Y. (2009). Stability of peakons for the Degasperis–Procesi equation, Commun. Pure Appl. Math. 62, 151–163. Liu, Y. and Yin, Z.W. (2006). Global existence and blow-up phenomena for the Degasperis–Procesi equation, Commun. Math. Phys. 267, 801–820. Logan, J.D. (1994). An Introduction to Nonlinear Partial Differential Equations, Wiley-Interscience, New York. Logan, J.D. and Dunbar, S.R. (1992). Travelling waves in model reacting flows with reversible kinetics, IMA J. Appl. Math. 49, 103–121.
Bibliography
833
Logan, J.D. and Shores, T.S. (1993a). Steady state solutions in a model reacting flow problem, Appl. Anal. 48, 273–286. Logan, J.D. and Shores, T.S. (1993b). On a system of nonlinear hyperbolic conservation laws with sources, Math. Models Methods Appl. Sci. 3, 341–358. Longuet-Higgins, M.S. (1974). On mass, momentum, energy and a circulation of a solitary wave, Proc. R. Soc. Lond. A337, 1–13. Longuet-Higgins, M.S. (1978a). The instability of gravity waves of finite amplitude in deep water I, Superharmonics, Proc. R. Soc. Lond. A360, 471–488. Longuet-Higgins, M.S. (1978b). The instabilities of gravity waves of finite amplitude in deep water II, Subharmonics, Proc. R. Soc. Lond. A360, 489–505. Longuet-Higgins, M.S. and Cocklet, E.D. (1976). The deformation of steep surface waves on water I. A numerical method of computation, Proc. R. Soc. Lond. A350, 1–26. Longuet-Higgins, M.S. and Fox, M.J.H. (1977). Theory of the almost highest wave: The inner solution, J. Fluid Mech. 80, 721–742. Lorenz, E.N. (1963a). Deterministic nonperiodic flow, J. Atmos. Sci. 20, 130–141. Lorenz, E.N. (1963b). The predictability of hydrodynamic flow, Trans. N. Y. Acad. Sci. Ser. II 25, 409–432. Lorenz, E.N. (1963c). The mechanics of vacillation, J. Atmos. Sci. 20, 448–464. Lorenz, E.N. (1964). The problem of deducing the climate from the governing equations, Tellus 16, 1–11. Lou, D.H. (2000). Planetary-scale baroclinic envelope Rossby solitons in a two-layer model and their interaction with synoptic-scale eddies, Dyn. Atmos. Oceans 32(1), 27–74. Ludu, A. and Draayer, J.P. (1998). Patterns on liquid surfaces: Cnoidal waves, compactons and scaling, Physica D123, 82–91. Ludu, A., Stoitcheva, G. and Draayer, J.P. (2000). Similarity analysis of nonlinear equations and bases of finite wavelength solitons, Int. J. Mod. Phys. E9, 263– 278. Luke, J.C. (1966). A perturbation method for nonlinear dispersive wave problems, Proc. R. Soc. Lond. A292, 403–412. Luke, J.C. (1967). A variational principle for a fluid with a free surface, J. Fluid Mech. 27, 395–397. Lundmark, H. (2007). Formation and dynamics of shock waves in the Degasperis– Procesi equation, J. Nonlinear Sci. 17, 169–198. Lundmark, H. and Szmigielski, J. (2003). Multi-peakon solutions of the Degasperis– Procesi equation, Inverse Probl. 9, 1241–1245. Ma, Y.-C. (1979). The perturbed plane-wave solution of the cubic Schrödinger equation, Stud. Appl. Math. 60, 43–58. Madsen, P.A., Murray, R. and Sorensen, O.R. (1991). A new form of the Boussinesq equations with improved linear dispersion characteristics, Coast. Eng. 15, 371– 388. Madsen, P.A. and Sorensen, O.R. (1992). A new form of the Boussinesq equations with improved linear dispersion characteristics. Part 2. A slowly varying bathymetry, Coast. Eng. 18, 183–204.
834
Bibliography
Madsen, P.A. and Sorensen, O.R. (1993). Bound waves and triad interactions in shallow water, Ocean Eng. 20, 359–388. Mainardi, F. (1994). On the initial-value problem for the fractional diffusion–wave equation, in Waves and Stability in Continuous Media (eds. S. Rionero, and T. Ruggeri), World Scientific, Singapore, 246–251. Mainardi, F. (1995). The time fractional diffusion–wave equation, Radiofisika 38, 20–36. Mainardi, F. (1996a). Fractional relaxation–oscillation and fractional diffusion–wave phenomena, Chaos Solitons Fractals 7, 1461–1477. Mainardi, F. (1996b). The fundamental solitons for the fractional diffusion–wave equations, Appl. Math. Lett. 9, 23–28. Makhankov, V.G. (1978). Dynamics of classical solitons, Phys. Rep. 35, 1–128. Manoranjan, V.S. and Mitchell, A.R. (1983). A numerical study of the Belousov– Zhabotinskii reaction using Galerkin finite element methods, J. Math. Biol. 16, 251–260. Marichev, O.I. (1983). Handbook of Integral Transforms of Higher Transcendental Functions, Theory and Algorithmic Table, Ellis Hoorwood, West Sussex. Maruno, K. and Biondini, G. (2004). Resonance and web structure in discrete soliton systems: the two-dimensional Toda lattice and its fully discrete and ultra-discrete analogue, J. Phys. A, Math. Gen. 37, 11819–11839. Maslow, S.A. (1986). Critical layers in shear flows, Annu. Rev. Fluid Mech. 18, 405– 432. Matsuno, Y. (2005a). Multisoliton solutions of the Degasperis–Procesi equation and their peakon limit, Inverse Probl. 21, 1553–1570. Matsuno, Y. (2005b). The N -soliton solution of the Degasperis–Procesi equation, Inverse Probl. 21, 2085–2101. Maxon, S. and Viecelli, J. (1974). Cylindrical solitons, Phys. Fluids 17, 1614–1616. McCall, S.L. and Hahn, E.L. (1967). Self-induced transparency by pulsed coherent light, Phys. Rev. Lett. 18, 908–911. McCall, S.L. and Hahn, E.L. (1969) Self-induced transparency, Phys. Rev. 183, 457– 485. McKean, H.P. (1970). Nagumo’s equation, Adv. Math. 4, 209–223. McKean, H.P. (1975). Application of Brownian motion to the equations of Kolmogorov–Pertovskii–Piskunov, Commun. Pure Appl. Math. 28, 323–331. McKean, H.P. (2003). Fredholm determinants and the Camassa–Holm hierarchy, Commun. Pure Appl. Math. 56, 638–680. McKean, H.P. (2004). Breakdown of the Camassa–Holm equation, Commun. Pure Appl. Math. 57, 416–418. Medina, E. (2002). An N soliton resonance solution for the KP equation: interaction with change of form and velocity, Lett. Math. Phys. 62, 91–99. Mei, C.C. (1989). The Applied Dynamics of Ocean Surface Waves, World Scientific, Singapore. Mikhailov, A.V. and Novikov, V.S. (2002). Perturbative symmetry approach, J. Phys. A, Math. Gen. 35, 4775–4790. Miles, J.W. (1980). Solitary waves, Annu. Rev. Fluid Mech. 12, 11–43.
Bibliography
835
Miles, J.W. (1981). The Korteweg–de Vries equation: A historical essay, J. Fluid Mech. 106, 131–147. Miura, R.M. (1968). Korteweg de Vries equation and generalizations. I. Remarkable explicit transformation, J. Math. Phys. 9, 1202–1204. Miura, R.M. (1974). Conservation laws for the fully nonlinear long wave equations, Stud. Appl. Math. L111, 45–56. Miura, R.M., Gardner, C.S., and Kruskal, M.D. (1968). Korteweg–de Vries equations and generalizations, II; Existence of conservation laws and constants of motion, J. Math. Phys. 9, 1204–1209. Miwa, T., Jimbo, M., and Date, E. (2000). Solitons: Differential Equations, Symmetries and Infinite-Dimensional Algebras, Cambridge University Press, Cambridge. Molinet, L. (2004). On well-posedness results for the Camassa–Holm equation on the line: a survey, J. Nonlinear Math. Phys. 11, 521–533. Mollenauer, L.F., Gordon, J.P., and Islam, M.N. (1986). Soliton propagation in long fibres with periodically compensated loss, IEEE J. Quantum Electron. 22, 1284– 1286. Montroll, E.W. and West, B.J. (1973). Models of population growth, diffusion, competition and rearrangement, in Synergetic (ed. H. Haken), B.G. Teubner, Stuttgart, 143–156. Munier, A., Burgen, J.R., Gutierrez, J., Fijalkow, E., and Feix, M.R. (1981). Group transformations and the nonlinear heat diffusion equation, SIAM J. Appl. Math. 40, 191–207. Murray, J.D. (1968). Singular perturbations of a class of nonlinear hyperbolic and parabolic equations, J. Math. Phys. 47, 111–133. Murray, J.D. (1970a). Perturbation effects on the decay of discontinuous solutions of nonlinear first order wave equations, SIAM J. Appl. Math. 19, 273–298. Murray, J.D. (1970b). On the Gunn-effect and other physical examples of perturbed conservation equations, J. Fluid Mech. 44, 315–346. Murray, J.D. (1973). On Burgers’ model equation for turbulence, J. Fluid Mech. 59, 263–279. Murray, J.D. (1993). Mathematical Biology, 2nd corrected edition, Springer, Berlin. Murray, J.D. (1997). Nonlinear-Differential Equation Models in Biology, Oxford University Press, Oxford. Myint-U, T. and Debnath, L. (2007). Linear Partial Differential Equations for Scientists and Engineers, 4th edition, Birkhäuser, Boston. Nagumo, J.S., Arimoto, S., and Yoshizawa, S. (1962). An active pulse transmission line simulating nerve axon, Proc. Inst. Radio Eng. 50, 2061–2071. Nangolnykh, K.A. and Ostrovsky, L.A. (1990). Nonlinear Wave Process in Acoustics, Nauka, Moscow. Naumkin, P.I. and Shishmarev, I.A. (1994). Nonlinear Nonlocal Equations in the Theory of Waves, Vol. 133, Am. Math. Soc., Providence. Nayfeh, A.H. (1965a). A perturbation method for treating nonlinear oscillation problems, J. Math. Phys. 44, 368–374. Nayfeh, A.H. (1965b). Nonlinear oscillations in a hot electron plasma, Phys. Fluids 8, 1896–1898.
836
Bibliography
Nayfeh, A.H. (1971). Third-harmonic resonance in the interaction of capillary and gravity waves, J. Fluid Mech. 48, 385–395. Nayfeh, A.H. (1973). Perturbation Methods, Wiley, New York. Nayfeh, A.H. and Hassan, S.D. (1971). The method of multiple scales and nonlinear dispersive waves, J. Fluid Mech. 48, 463–475. Nayfeh, A.H. and Saric, W.H. (1971). Nonlinear Kelvin–Helmholtz instability, J. Fluid Mech. 55, 311–327. Newell, A.C. (1980). The inverse scattering transform, in Topics in Current Physics (eds. R. Bullough, and P. Caudrey), Vol. 17, Springer, Berlin, 177–242. Newell, A.C. (1985). Solitons in Mathematics and Physics, SIAM, Philadelphia. Newell, A.C. and Kaup, D.J. (1978). Solitons as particles, oscillations, and in slowly changing media: A singular perturbation theory, Proc. R. Soc. Lond. A361, 413– 446. Newell, A.C. and Moloney, J.V. (1992). Nonlinear Optics, Addison-Wesley, New York. Newman, W.I. (1980). Some exact solutions to a nonlinear diffusion problem in population genetics and combustion, J. Theor. Biol. 85, 325–334. Newman, W.I. (1983). Nonlinear diffusion: Self-similarity and travelling waves, Pure Appl. Geophys. 121, 417–441. Nigmatullin, R.R. (1986). The realization of the generalized transfer equation in a medium with fractal geometry, Phys. Status Solidi (b) 133, 425–430. Nimmo, J.J.C. and Freeman, N.C. (1983). The use of Bäcklund transformations in obtaining N -soliton solutions in Wronskian form, J. Phys. A, Math. Gen. 17, 1415–1424. Novick-Cohen, A. and Sivashinsky, G.I. (1986). On the solidification front of a dilute binary alloy: Thermal diffusivity effects and breathing solutions, Physica D20, 237–258. Nozaki, K. and Taniuti, T. (1973). Propagation of solitary pulses in interactions of plasma waves, J. Phys. Soc. Jpn. 34, 796–800. Nunziato, J.W., Walsh, E.K., Schuler, K.W., and Baker, L.M. (1974). Wave propagation in nonlinear viscoelastic solids in Encyclopedia of Physics (ed. S. Flügge), Vol. V1a/4, Springer, Berlin. Nwogu, O. (1993). Alternative form of Boussinesq equations for nearshore wave propagation, J. Waterw., Port, Coastal, Ocean Eng., ASCE 119, 618–638. Nye, J.F. (1960). The response of glaciers and ice-sheets to seasonal and climatic changes, Proc. R. Soc. Lond. A256, 559–584. Nye, J.F. (1963). The response of a glacier to changes in the rate of nourishment and wastage, Proc. R. Soc. Lond. A275, 87–112. Oberhettinger, F. (1972). Tables of Bessel Transforms, Springer, New York. Ohta, Y., Maruno, K.I., and Feng, B.F. (2008). An integrable semi-discretization of Camassa–Holm equation and its determinant solution, J. Phys. A, Math. Theor. 41, 1–30. Oikawa, M. and Yajima, N. (1973). Interactions of solitary waves—A perturbation approach to nonlinear systems, J. Phys. Soc. Jpn. 34, 1093–1099. Oikawa, M. and Yajima, N. (1974a). A class of exactly solvable nonlinear evolution equations, Prog. Theor. Phys. Suppl. 54, 1576–1577.
Bibliography
837
Oikawa, M. and Yajima, M. (1974b). A perturbation approach to nonlinear systems II. Interaction of nonlinear modulated waves, J. Phys. Soc. Jpn. 37, 486–496. Okubo, A. (1980). Diffusion and Ecological Problems: Mathematical Models, Springer, Berlin. Olver, P.J. (1984a). Hamiltonian and non-Hamiltonian models for water waves, in Trends and Applications of Pure Mathematics to Mechanics (eds. P.G. Ciarlet, and M. Roseau), Lecture Notes in Physics, Vol. 195, Springer, New York. Olver, P.J. (1984b). Hamiltonian perturbation theory and water waves, Contemp. Math. 28, 231–249. Olver, P.J. (1993). Applications of Lie Groups to Differential Equations, 2nd edition, Springer, New York. Ono, H. (1975). Algebraic solitary waves in stratified fluids, J. Phys. Soc. Jpn. 39, 1082–1091. Oron, A. and Rosenau, P. (1989). Evolution of the coupled Bérnard-Marangoni convection, Phys. Rev. A39, 2063–2069. Ostrovsky, L.A. (1976). Short-wave asymptotics for weak shock waves and solitons in mechanics, Int. J. Non-Linear Mech. 11, 401–406. Ostrovsky, L.A. and Pelinovsky, E.N. (1971). Averaging method for nonsinusoidal waves, Sov. Phys. Dokl. 15, 1097–1099. Ostrovsky, L.A. and Pelinovsky, E.N. (1972). Method of averaging and the generalized variational principle for nonsinusoidal wave, J. Appl. Math. Mech. (Prikl. Math. Mech.) 36, 63–78. Otto, E. and Sudan, R.N. (1971). Nonlinear theory of ion acoustic waves with Landeu damping, Phys. Fluids 13, 2388–2394. Pack, D.C. (1960). A note on the breakdown of continuity in the motion of a compressible fluid, J. Fluid Mech. 8, 103–108. Page, B. (1990). Asymptotic solutions for localized vibrational modes in strongly anharmonic periodic systems, Phys. Rev. B41, 7835–7837. Palais, R.S. (1997). The symmetrics of solitons, Bull. Am. Math. Soc. 34, 339–403. Pao, C.V. (1992). Nonlinear Elliptic and Parabolic Equations, Plenum, New York. Parau, E.I., Vanden-Broeck, J.M., and Cooker, M.J. (2005). Nonlinear threedimensional gravity-capillary solitary waves, J. Fluid Mech. 536, 99–105. Parker, D.F. (1980). The decay of saw-tooth solutions to the Burgers equation, Proc. R. Soc. Lond. A369, 409–424. Parker, A. (2004). On the Camassa–Holm equation and a direct method of solution: I. Bilinear form and solitary waves, Proc. R. Soc. Lond. 460, 2929–2957. Parker, A. (2005a). On the Camassa–Holm equation and a direct method of solution: II. Soliton solutions, Proc. R. Soc. Lond. 461, 3611–3632. Parker, A. (2005b). On the Camassa–Holm equation and a direct method of solution: III. N -soliton solutions, Proc. R. Soc. Lond. 461, 3893–3911. Pattle, R.E. (1959). Diffusion from an instantaneous point source with a concentration-dependent coefficient, Q. J. Mech. Appl. Math. 12, 407–409. Pawlik, M. and Rowlands, G. (1975). The propagation of solitary waves in piezoelectric semiconductors, J. Phys. C8, 1189–1204. Peipman, T., Valdek, U., and Engelbrecht, J. (1992). Nonlinear two-dimensional longitudinal and shear waves in solids, Acoustica 76, 84–94.
838
Bibliography
Pelinovsky, D. and Grimshaw, R. (1997). Structural transformation of eigenvalues for a perturbed algebraic soliton potential, Phys. Lett. A 229, 165–172. Pelinovsky, E., Polukhina, O., Slunyaev, A., and Talipova, T. (2007). Internal Solitary Waves, in Solitary Waves in Fluids (ed. R.H.J. Grimshaw), 85–110. Penel, P. and Brauner, C.M. (1974). Identification of parameters in a nonlinear selfconsistent system including a Burgers equation, J. Math. Anal. Appl. 45, 654– 681. Peregrine, D.H. (1967). Long waves on a beach, J. Fluid Mech. 27, 815–827. Peregrine, D.H. (1983). Water waves, nonlinear Schrödinger equations and their solutions, J. Aust. Math. Soc. B25, 16–43. Perring, J.K. and Skyrme, T.H.R. (1962). A model of unified field equation, Nucl. Phys. 31, 550–555. Phillips, O.M. (1981). Wave interactions—the evolution of an idea, J. Fluid Mech. 106, 215–227. Plotnikov, P.I. and Tolland, J.F. (2004). Convexity of Stokes waves of extreme form, Arch. Ration. Mech. Anal. 17, 349–416. Podlubny, I. (1999). Fractional Differential Equations, Vol. 198, Academic Press, Boston. Prasad, P. (1973). Nonlinear wave propagation on an arbitrary steady transonic flow, J. Fluid Mech. 57, 721–737. Prasad, P. (1975). Approximation of perturbation equations of a quasi-linear hyperbolic system in a neighbourhood of a bicharacteristic, J. Math. Anal. Appl. 50, 470–482. Prasad, P. and Ravindran, R. (1977). A theory of nonlinear waves in multidimensions: with special reference to surface water waves, J. Inst. Math. Appl. 20, 9–20. Pullin, D.I. and Grimshaw, R.H.J. (1988). Finite amplitude solitary waves at the interface between two homogeneous fluids, Phys. Fluids 31, 3550–3559. Qiao, Z. (1995). Generation of the hierarchies of solitons and generalized structure of the commutator representation, Acta Math. Appl. Sin. 18, 287–301. Qiao, Z. (2001). Generalized matrix structure and algebro-geometric solution for integrable systems, Rev. Math. Phys. 13, 545–586. Qiao, Z. (2002). Finite-dimensional Integrable Systems and Nonlinear Evolution Equations, Chinese Higher Education Press, Beijing. Qiao, Z. (2003). The Camassa–Holm hierarchy, related N -dimensional integrable systems and algebro-geometric solution on a symplectic submanifold, Commun. Math. Phys. 239, 309–341. Qiao, Z.J. (2006). A new integrable equation with cuspons and W/M-shape-peaks solitons, J. Math. Phys. 47, 112701–112709. Qiao, Z.J. and Li, S. (2004). A new integrable hierarchy, parametric solution, and traveling wave solution, Math. Phys. Anal. Geom. 7, 289–308. Qiao, Z. and Strampp, W. (2002). Negative order MKdV hierarchy and a new integrable Neumann-like system, Physica A 313, 365–380. Qiao, Z.J. and Zhang, G. (2006). On peaked and smooth solitons for the Camassa– Holm equation, Europhys. Lett. 73, 657–663.
Bibliography
839
Ramasami, E.K. and Debnath, L. (1994). Similarity solutions of the nonlinear Schrödinger equation, J. Math. Anal. Appl. 187, 851–863. Ravindran, R. and Prasad, P. (1979). A mathematical analysis of nonlinear waves in a fluid filled viscoelastic tube, Acta Mech. 31, 253–280. Rayleigh, L. (1876). On waves, Philos. Mag. 1, 257–279. Redekopp, L.G. and Weidman, P.D. (1968). Solitary Rossby waves in zonal shear flows and their interactions, J. Atmos. Sci. 35, 790–804. Riabouchinsky, D. (1932). Sur l’analogie hydraulique des mouvements d’un fluide compressible, Acad. Sci., Comptes Rendus 195, 998–1015. Richards, P.I. (1956). Shock waves on the highway, Oper. Res. 4, 42–51. Riemann, B., (1858). Uber die Fortpflanzung ebener Luftwellen von endlicher Schwingungsweite, Göttingen Abhandlunger, Vol. viii, p. 43 (Werke, 2te Aufl., Leipzig, 1892, p. 157). Rodin, E.Y. (1970). On some approximate and exact solutions of boundary value problems for Burgers equations, J. Math. Anal. Appl. 30, 401–414. Rodriguez-Blanco, G. (2001). On the Cauchy problem for the Camassa–Holm equation, Nonlinear Anal. 46, 309–327. Rogers, C. and Ames, W.F. (1989). Nonlinear Boundary Value Problems in Science and Engineering, Academic Press, New York. Rosales, R.R. (1978). The similarity solution for the Korteweg–de Vries equation and the related Painlevé transcendent, Proc. R. Soc. Lond. A361, 265–275. Rosenau, P. (1982). A nonlinear thermal wave in a reacting medium, Physica 5D, 136–144. Rosenau, P. (1994). Nonlinear dispersion and compact structures, Phys. Rev. Lett. 73, 1737–1741. Rosenau, P. (1997). On nonanalytical solitary waves formed by a nonlinear dispersion, Phys. Lett. A230, 305–318. Rosenau, P. (2000). Compact and noncompact dispersive patterns, Phys. Lett. A275, 193–198. Rosenau, P. and Hyman, J.M. (1993). Compactons: Solitons with finite wavelength, Phys. Rev. Lett. 70, 564–567. Rosenau, P. and Levy, D. (1999). Compactons in a class of nonlinearity quintic equations, Phys. Lett. A252, 297–304. Rowland, H.L., Lyon, J.G., and Papadopoulus, K. (1981). Strong Langmuir turbulence in one and two dimensions, Phys. Rev. Lett. 46, 346–349. Rudenko, O.V. and Soluyan, S.I. (1977). Theoretical Foundations of Nonlinear Acoustics, Consultants Bureau (Plenum), New York. (English translation by R.T. Beyer.) Russell, J.S., (1844). Report on waves, Report on the 14th Meeting of the British Association for the Advancement of Science (ed. J. Murray), London, 311–390. Sabini, K. and Serebryany, A. (2005). Intense short-period internal waves in the ocean, J. Mar. Resour. 63, 227–261. Sachdev, P.L. (1987). Nonlinear Diffusive Waves, Cambridge University Press, Cambridge. Sagdeev, R.Z. (1966). Review of Plasma Physics Vol. IV, Consultants Bureau, New York (ed. M.A. Leontovich).
840
Bibliography
Sandri, G. (1963). The foundations of nonequilibrium statistical mechanics I, II, Ann. Phys. 24, 332–379. Sandri, G. (1965). A new method of expansion in mathematical physics—I, Nuovo Cimento B36, 67–93. Sandri, G. (1967). Uniformization of asymptotic expansions, in Nonlinear Partial Differential Equations: In A Symposium on Methods of Solutions (ed. W.F. Ames), Academic Press, New York, 259–277. Sandusky, K.W., Page, J.B., and Schmidt, K.E. (1992). Stability and motion of intrinsic localized modes in nonlinear periodic lattices, Phys. Rev. B46, 6161–6170. Sanuki, H., Shimizu, K., and Todoroki, J. (1972). Effects of Landau damping on nonlinear wave modulation in plasma, J. Phys. Soc. Jpn. 33, 198–205. Satsuma, J. (1987a). Topics in Soliton Theory and Exactly Solvable Nonlinear Equations, World Publishing Company, Singapore (eds. M.A. Ablowitz, et al.). Satsuma, J. (1987b). Explicit solutions of nonlinear equations with density dependent diffusion, J. Phys. Soc. Jpn. 56, 1947–1950. Schiff, L.I. (1951). Nonlinear meson theory of nuclear forces, Phys. Rev. 84, 1–11. Schimizu, K. and Ichikawa, V.H. (1972). Automodulation of ion oscillation modes in plasma, J. Philos. Soc. Jpn. 33, 789–792. Schmidt, G. (1975). Stability of envelope solitons, Phys. Rev. Lett. 34, 724–726. Schneider, W.R. and Wyss, W. (1989). Fractional diffusion and wave equations, J. Math. Phys. 30, 134–144. Scott, A.C. (1969). A nonlinear Klein–Gordon equation, Am. J. Phys. 37, 52–61. Scott, A.C. (1970). Active and Nonlinear Wave Propagation in Electronics, WileyInterscience, New York. Scott, A.C. (1977). Neurophysics, Wiley, New York. Scott, A.C. (2003). Nonlinear Science, 2nd edition, Oxford University Press, Oxford. Scott, A.C., Chu, F.Y.F., and McLaughlin, D.W. (1973). The soliton—A new concept in applied science, Proc. IEEE 61, 1443–1483. Seddon, J.A. (1900). River hydraulics, Trans. Am. Soc. Civ. Eng. 43, 179–243. Segur, H. (1973). The Korteweg–de Vries equation and water waves. I. Solutions of the equation, J. Fluid Mech. 59, 721–736. Seliger, R.L. (1968). On the breaking of waves, Proc. R. Soc. Lond. A303, 493–496. Sen, A.R. (1963). Surface waves due to blasts on and above liquids, J. Fluid Mech. 16, 65–81. Serre, D. (2000). Systems of Conservation Laws, II, Cambridge University Press, Cambridge. Sharma, V.D. (2010). Quasilinear Hyperbolic Systems, Compressible Flows and Waves, CRC Press, Boca Raton. Shigesada, N. (1980). Spatial distribution of dispersing animals, J. Math. Biol. 9, 85–96. Shimizu, K. and Ichikawa, V.H. (1972). Automodulation of ion oscillation modes in plasma, J. Phys. Soc. Jpn. 33, 789–792. Shinbrot, M. (1970). The shallow water equations, J. Eng. Math. 4, 293–304. Shinbrot, M. (1976). The initial-value problem for surface waves under gravity, I: The simplest case, Indiana Univ. Math. J. 25, 281–300.
Bibliography
841
Shinbrot, M. and Friedman, A. (1968). The initial-value problem for the linearized equations of water waves, J. Math. Mech. 18, 1177–1193. Shivamoggi, B.K. and Debnath, L. (1983). Modulational stability of Korteweg– de Vries and Boussinesq wavetrains, Int. J. Math. Math. Sci. 6, 811–817. Short, M. (1955). The initiation of detonation from general non-uniformly distributed initial conditions, Philos. Trans. R. Soc. Lond. A353, 173–203. Showalter, K. and Tyson, J.J. (1987). Luther’s 1906 discovery and analysis of chemical waves, J. Chem. Educ. 64, 742–744. Sievers, A.J. and Takeno, S. (1988). Intrinsic localized modes in anharmonic crystals, Phys. Rev. Lett. 61, 970–973. Simmen, J.A. and Saffman, P.G. (1985). Steady deep water waves on a linear shear current, Stud. Appl. Math. 75, 35–57. Skyrme, T.H.R. (1958). A nonlinear theory of strong interactions, Proc. R. Soc. Lond. A247, 260–278. Skyrme, T.H.R. (1961). Particle states of a quantized meson field, Proc. R. Soc. Lond. A262, 237–245. Sleeman, B.D. (1982). Small amplitude periodic waves for the FitzHugh–Nagumo equation, J. Math. Biol. 14, 309–325. Smoller, J. (1994). Shock Waves and Reaction–Diffusion Equations, 2nd edition, Springer, New York. Sneddon, I.N. (1951). Fourier Transforms, McGraw-Hill, New York. Soewono, E. and Debnath, L. (1990). Asymptotic stability of solutions of the generalized Burgers equation, J. Math. Anal. Appl. 153, 179–189. Soewono, E. and Debnath, L. (1994). Classification of self-similar solutions to a generalized Burgers equation, J. Math. Anal. Appl. 184, 389–398. Sparrow, E.M., Quack, H., and Boerner, C.J. (1970). Local nonsimilarity boundarylayer solutions, AIAA J. 8, 1342–1350. Stakgold, I. (1971). Branching of solutions on nonlinear equation, SIAM Rev. 13, 289–332. Stakgold, I. (1979). Green’s Functions and Boundary Value Problems, WileyInterscience, New York. Stakgold, I. and Payne, L.E. (1973). Nonlinear problems in nuclear reactor analysis, in Nonlinear Problems in the Physical Sciences and Biology, Lecture Notes in Mathematics, Vol. 322, Springer, New York, 298–307. Stewartson, K. and Stuart, J.T. (1971). A nonlinear instability theory for wave system in plane Poiseuille flow, J. Fluid Mech. 48, 529–545. Stoker, J.J. (1957). Water Waves, Interscience, New York. Stokes, G. (1847). On the theory of oscillatory waves, Trans. Camb. Philos. Soc. 8, 197–229. Strauss, W. (2010). Steady Water Waves, Bull. Am. Math. Soc. 47, 2. Struik, D.J. (1926). Détermination rigoureuse des ondes irrotationnelles périodiques dans un canal á profondeur finie, Math. Ann. 95, 595–634. Stuart, J.T. (1960). On the nonlinear mechanics of wave disturbances in stable and unstable parallel flows, Part 1. The basic behaviour in plane Poiseuille flow, J. Fluid Mech. 9, 353–370.
842
Bibliography
Stuart, J.T. and DiPrima, R.C. (1978). The Eckhaus and Benjamin–Feir resonance mechanisms, Proc. R. Soc. Lond. A362, 27–41. Sturrock, P.A. (1957). Nonlinear effects in electron plasma, Proc. R. Soc. Lond. A242, 277–299. Su, C.S. and Gardner, C.S. (1969). The Korteweg–de Vries equation and generalizations, III. Derivation of the Korteweg–de Vries equation and Burgers’ equation, J. Math. Phys. 10, 536–539. Sun, S.M. (1999). Non-existence of truly solitary waves in water with small surface tension, Proc. R. Soc. Lond. A455, 2191–2228. Talanov, V.I. (1965). Self-focusing of wave beams in nonlinear media, JETP Lett. 2, 138–141. Tang, S. and Webber, R.O. (1991). Numerical study of Fisher’s equation by a Petrov– Galerkin finite element method, J. Aust. Math. Soc. B33, 27–38. Taniuti, T. (1974). Reductive perturbation method and far fields of wave equations, Prog. Theor. Phys. Suppl. 55, 1–35. Taniuti, T. and Nishihara, K. (1983). Nonlinear Waves, Pitman Advanced Publishing Program, Boston. Taniuti, T. and Washimi, H. (1968). Self-trapping and instability of hydromagnetic waves along the magnetic field in a cold plasma, Phys. Rev. Lett. 21, 209–212. Taniuti, T. and Wei, C.C. (1968). Reductive perturbation method in nonlinear wave propagation-I, J. Phys. Soc. Jpn. 24, 941–946. Taniuti, T. and Yajima, N. (1969). Perturbation method for a nonlinear wave modulation I, J. Math. Phys. 10, 1369–1372. Taniuti, T. and Yajima, N. (1973). Perturbation method for a nonlinear wave modulation III, J. Math. Phys. 14, 1389–1397. Tao, T. (2002). Low-regularity global solutions to nonlinear dispersive equations, Surveys in Analysis and Operator Theory, Proc. Centre Math. Appl, Austral. Nat. Univ., 19–48. Tao, T. (2006). Nonlinear Dispersive Equations, Local and Global Analysis, Am. Math. Soc., Providence. Tappert, F. and Varma, C.M. (1970). Asymptotic theory of self-trapping of heat pulses in solids, Phys. Rev. Lett. 25, 1108–1111. Teles Da Silva, A.F. and Peregrine, D.H. (1988). Steep solitary waves in water of finite depth with constant vorticity, J. Fluid Mech. 195, 281–305. Thyagaraja, A. (1979). Recurrent motions in certain continuum dynamical system, Phys. Fluids 22, 2093–2096. Thyagaraja, A. (1981). Recurrence, dimensionality, and Lagrange stability of solutions of the nonlinear Schödinger equation, Phys. Fluids 24, 1973–1975. Thyagaraja, A. (1983). Recurrence phenomena and the number of effective degrees of freedom in nonlinear wave motions, in Nonlinear Waves (ed. L. Debnath), Cambridge University Press, Cambridge, 308–325. Tien-Yien, L. and Yorke, J.A. (1975). Period three implies chaos, Am. Math. Mon. 82(10), 985–992. Toda, M. (1967a). Vibration of a chain with nonlinear interaction, J. Phys. Soc. Jpn. 22, 431–436.
Bibliography
843
Toda, M. (1967b). Wave propagation in anharmonic lattices, J. Phys. Soc. Jpn. 23, 501–506. Toda, M. (1975). Studies of a Nonlinear Lattice, Phys. Rep. 18, 1–123. Toda, M. and Wadati, M. (1973). A soliton and two solitons in an exponential lattice and related equations, J. Phys. Soc. Jpn. 34, 18–25. Toland, J.F. (1978). On the existence of a wave of greatest height and Stokes’ conjecture, Proc. R. Soc. Lond. A363, 469–485. Turin, A.M. (1952). The chemical basis of morphogenesis, Philos. Trans. R. Soc. Lond. B237, 37–72. Vakhnenko, V. and Parkes, E. (2004). Periodic and solitary-wave solutions of the Degasperis-Procesi equation, Chaos Solitons Fractals 20, 1059–1073. Valenti, A. (1988). The asymptotic analysis of nonlinear waves in magneto-thermoelastic solids with thermal relaxation, ZAMP 39, 299–312. Vanden-Broeck, J.M. (1994). Steep solitary waves in water of finite depth with constant vorticity, J. Fluid Mech. 274, 339–348. Vanden-Broeck, J.M. (1995). New families of steep solitary waves in water of finite depth with constant vorticity, Eur. J. Mech. B, Fluids 14, 761–774. Vanden-Broeck, J.M. (1996a). Periodic waves with constant vorticity in water of infinite depth, IMA J. Appl. Math. 56, 207–217. Vanden-Broeck, J.M. (1996b). Numerical calculations of the free-surface flow under a sluice gate, J. Fluid Mech. 330, 339–347. Vanden-Broeck, J.M. (2004). Nonlinear capillary free-surface flows, J. Eng. Math. 50, 415–426. Vanden-Broeck, J.M. (2007). Solitary Waves in Water: Numerical Methods and Results, in Solitary Waves in Fluids (ed. R.H.J. Grimshaw), 55–84. Vanden-Broeck, J.M. (2010). Gravity-Capillary Free-Surface Flows, Cambridge University Press, London. Vanden-Broeck, J.M. and Dias, F. (1992). Gravity-capillary solitary waves in water of infinite depth and related free-surface flows, J. Fluid Mech. 240, 549–557. Van Saarloos, W. (1989). Front propagation into unstable states: Marginal stability as a dynamical mechanism for velocity selection, Phys. Rev. A 37, 211–229. Van Wijngaarden, L. (1968). On the equation of motion for mixtures of liquid and gas bubble, J. Fluid Mech. 33, 465–474. Wadati, M. (1972). The exact solution of the modified Korteweg–de Vries equation, J. Phys. Soc. Jpn. 32, 1681–1687. Wadati, M. (1973). The modified Korteweg–de Vries equation, J. Phys. Soc. Jpn. 34, 1289–1296. Wadati, M. and Toda, M. (1972). The exact N -soliton solution of the Korteweg– de Vries equation, J. Phys. Soc. Jpn. 32, 1403–1411. Wadati, M., Ichikawa, Y.H., and Shimizu, T. (1980). Cusp soliton of a new integrable evolution equation, Prog. Theor. Phys. 64, 1959–1967. Wadati, M., Konno, K., and Ichikawa, Y.H. (1979). New integrable nonlinear evolution equations, J. Phys. Soc. Jpn. 47, 1698–1700. Wahlquist, H.D. and Estabrook, F.B. (1973). Bäcklund transformations for solutions of the Korteweg–de Vries equation, Phys. Rev. Lett. 23, 1386–1389.
844
Bibliography
Wahlquist, H.D. and Estabrook, F.B. (1975). Prolongation structures and nonlinear evolution equations, J. Math. Phys. 16, 1–17. Wahlquist, H.D. and Estabrook, F.B. (1976). Prolongation structures and nonlinear evolution equations, J. Math. Phys. 17, 1403–1411. Walsh, R.A. (1969). Initial-value problems associated with ut (x, t) = δuxx (x, t) − u(x, t)ux (x, t), J. Math. Anal. Appl. 26, 235–247. Wang, X.Y. (1985). Nerve propagation and wall in liquid crystals, Phys. Lett. 112A, 402–406. Wang, X.Y. (1986). Brochard–Leger wall in liquid crystals, Phys. Rev. A34, 5179– 5182. Wang, X.Y., Shu, Z.S., and Lu, Y.K. (1990). Solitary wave solutions of the generalized Burgers–Huxley equation, J. Phys. Math. Gen. 23, 271–274. Ward, R.S. (1984). The Painlevé property for the self-dual gauge-field equations, Phys. Lett. 102A, 279–282. Ward, R.S. (1985). Integrable and solvable systems and relations among them, Philos. Trans. R. Soc. Lond. A315, 451–457. Ward, R.S. (1986). Multi-dimensional integrable systems, in Field Theory, Quantum Gravity, and Strings, II Proceedings, Meudon and Paris, VI, France 1985/1986 (eds. H.J. de Vega and N. Sanchez), Lect. Notes Phys., Vol. 280, Springer, Berlin, 106–110. Washimi, H. and Taniuti, T. (1966). Propagation of ion-acoustic solitary waves of small amplitude, Phys. Rev. Lett. 17, 996–998. Watson, J. (1960). On the nonlinear mechanics of wave disturbances in stable and unstable parallel flows, Part 2. The development of a solution for plane Poiseuille flow and for plane Couette flow, J. Fluid Mech. 9, 371–389. Wazwaz, A.M. (2002). Partial Differential Equation Methods and Applications, A.A. Balkema Publishers, Tokyo. Wehausen, J.V. and Laitone, E.V. (1960). Surface waves, in Handbuch der Physik (ed. W. Flügge), Vol. 9, Springer, Berlin, 446–778. Weidman, P.D. and Maxworthy, T. (1978). Experiments on strong interactions between solitary waves, J. Fluid Mech. 85, 417–431. Weiland, J. and Wiljelmsson, H. (1977). Coherent Nonlinear Interaction of Waves in Plasmas, Pergamon Press, Oxford. Weiland, J., Ichikawa, Y.H., and Wiljelmsson, H. (1978). A perturbation expansion for the NLS equation with application to the influence of nonlinear Landau damping, Phys. Scr. 17, 517–522. Whitham, G.B. (1956). On the propagation of weak shock waves, J. Fluid Mech. 1, 290–318. Whitham, G.B. (1965a). A general approach to linear and nonlinear dispersive waves using a Lagrangian, J. Fluid Mech. 22, 273–283. Whitham, G.B. (1965b). Nonlinear dispersive waves, Proc. R. Soc. Lond. A283, 238– 261. Whitham, G.B. (1967a). Nonlinear dispersion of water waves, J. Fluid Mech. 27, 399–412. Whitham, G.B. (1967b). Variational methods and application to water waves, Proc. R. Soc. Lond. A299, 6–25.
Bibliography
845
Whitham, G.B. (1970). Two-timing, variational principles and waves, J. Fluid Mech. 44, 373–395. Whitham, G.B. (1974). Linear and Nonlinear Waves, Wiley, New York. Whitham, G.B. (1975). A new approach to problems of shock dynamics. Part I. Two dimensional problems, J. Fluid Mech. 2, 146–171. Whitham, G.B. (1984). Comments on periodic waves and solitons, IMA J. Appl. Math. 32, 353–366. Witting, J.M. (1984). A unified model for the evolution of nonlinear water waves, J. Comput. Phys. 56, 203–236. Wyss, W. (1986). The fractional diffusion equation, J. Math. Phys. 27, 2782–2785. Yajima, N. and Outi, A. (1971). A new example of stable solitary waves, Prog. Theor. Phys. 45, 1997–2005. Yin, Z.Y. (2003a). Global existence for a new periodic integrable equation, J. Math. Anal. Appl. 283, 129–139. Yin, Z.Y. (2003b). On the Cauchy problem for an integrable equation with peakon solutions, Ill. J. Math. 47, 649–666. Yin, Z.Y. (2004a). On the blow-up scenario for the generalized Camassa–Holm equation, Commun. Partial Differ. Equ. 29, 867–877. Yin, Z.Y. (2004b). Global solutions to a new integrable equation with peakons, Indiana Univ. Math. J. 53, 1189–1210. Yorke, J. and Li, T.-Y. (1975). Period three implies chaos, Am. Math. Mon. 82(10), 985–992. Yuen, H.C. and Ferguson, W.E., Jr. (1978a). Relationship between Benjamin–Feir instability and recurrence in the nonlinear Schrödinger equation, Phys. Fluids 21, 1275–1278. Yuen, H.C. and Ferguson, W.E., Jr. (1978b). Fermi–Pasta–Ulam recurrence in the two space dimensional nonlinear Schrödinger equation, Phys. Fluids 21, 2116– 2118. Yuen, H.C. and Lake, B.M. (1975). Nonlinear deep water waves: Theory and experiment, Phys. Fluids 18, 956–960. Yuen, H.C. and Lake, B.M. (1979). A new model for nonlinear wind waves, Part 2, Theoretical, Stud. Appl. Math. 60, 261–270. Yuen, H.C. and Lake, B.M. (1980). Instabilities of wave on deep water, Annu. Rev. Fluid Mech. 12, 303–334. Yuen, H.C. and Lake, B.M. (1982). Nonlinear dynamics of deep-water gravity waves, Adv. Appl. Mech. 22, 67–229. Zabusky, N.J. (1967). A synergetic approach to problems of nonlinear dispersive wave propagation and interaction, in Proc. Symp. on Nonlinear Partial Differential Equations (ed. W.F. Ames), Academic Press, Boston. Zabusky, N.J. and Galvin, C.J. (1971). Shallow water waves, the Korteweg–de Vries equation and solitons, J. Fluid Mech. 47, 811–824. Zabusky, N.J. and Kruskal, M.D. (1965). Interaction of solitons in a collisionless plasma and the recurrence of initial states, Phys. Rev. Lett. 15, 240–243. Zakharov, V.E. (1968a). Stability of periodic waves of finite amplitude on the surface of a deep fluid, J. Appl. Mech. Tech. Phys. 9, 86–94.
846
Bibliography
Zakharov, V.E. (1968b). Stability of periodic waves of finite amplitude on the surface of a deep fluid, J. Appl. Mech. Tech. Phys. 2, 190–194. Zakharov, V.E. and Faddeev, L.D. (1971). Korteweg–de Vries equation, a completely integrable Hamiltonian system, Funct. Anal. Appl. 5, 280–287. Zakharov, V.E. and Shabat, A.B. (1972). Exact theory of two-dimensional self focusing and one-dimensional self modulation of waves in nonlinear media, Sov. Phys. JETP 34, 62–69. Zakharov, V.E. and Shabat, A.B. (1973). Interaction between solitons in a stable medium, Sov. Phys. JETP 37, 823–828. Zakharov, V.E. and Shabat, A.B. (1974). A scheme for integrating the nonlinear equations of mathematical physics by the method of the inverse scattering problem, Funct. Anal. Appl. 8, 226–235. Zakharov, V.E. and Shulman, E.I. (1982). To the integrability of the system of two coupled nonlinear Schrödinger equations, Physica 4D, 270–274. Zauderer, E. (1989). Partial Differential Equations of Applied Mathematics, 2nd edition, Wiley-Interscience, New York. Zel’dovich, Ya.B. (1980). Regime classification of an exothermic reaction with nonuniform initial conditions, Combust. Flame 39, 211–214. Zel’dovich, Ya.B. and Raizer, Yu.P. (1966). Physics of Shock Waves and High Temperature Hydrodynamic Phenomena, Academic Press, New York. Zel’dovich, Ya.B. and Raizer, Yu.P. (1968). Elements of Gas Dynamics and the Classical Theory of Shock Waves, Academic Press, New York. Zel’dovich, Ya.B., Barenblat, G.I., Librovich, V.B., and Makhviladzd, G.M. (1985). The Mathematical Theory of Combustion and Explosion, Consultants Bureau, Division Plenum Press, New York. Zel’dovich, Ya.B., Librovich, V.B., Makhviladze, G.M., and Sivashinsky, G.I. (1970). Development of detonation in a non-uniformly preheated gas, Astr. Acta 15, 313–321. Zhang, X. (1995). Capillary-gravity and capillary waves generated in a wind wave tank: observations and theories, J. Fluid Mech. 289, 51–82. Zhang, G. and Qiao, Z. (2007). Cuspons and smooth solitons of the Degasperis– Procesi equation under inhomogeneous boundary condition, Math. Phys. Anal. Geom. 10, 205–225. Zheng, Y. (2001). Systems of Conservation Laws: Two Dimensional Riemann Problems, Birkhäuser, Basel. Zhou, Y. (2004a). Wave breaking for a periodic shallow water equation, J. Math. Anal. Appl. 290, 591–604. Zhou, Y. (2004b). Wave breaking for a shallow water equation, Nonlinear Anal. 57, 137–152. Zhou, Y. (2005). Local well-posedness and blow-up criteria for a rod equation, Math. Nachr. 278, 1726–1739. Zhou, X. and Grimshaw, R. (1989). The effect of variable currents on internal solitary waves, Dyn. Atmos. Oceans 14, 17–39. Zufira, J.A. (1987). Symmetry breaking in periodic and solitary gravity-capillary waves on water of finite depth, J. Fluid Mech. 184, 183–206.
Index
Abdelkader, 417 Abel integral equation of the first kind, 765 Abel integral equation of the second kind, 766 Abel’s formula, 86 Ablowitz, 152, 155, 417, 425, 506, 542 Ablowitz and Ladik (AL) equation, 199 Absorption isotherm, 300 Acoustic radiation potential, 9, 65 Acrivos, 153 Action, 198 Action integral, 165 Addition theorems for elliptic functions, 717 Adiabatic invariant, 588 Airy, 427 Airy differential equation, 570 Airy function, 10, 38, 156, 468, 702 Airy integral, 38 AKNS method, 425, 506, 508 Alfvén wave velocity, 358 Amick, 182 Ammerman, 152, 397 Amplitude dispersion, 181 Amplitude parameter, 183 Angular frequencies, 24 Anharmonic forces, 515 Antikinks, 431 Antisoliton, 597, 600, 605 Aoki, 152, 397 Arecchi, 151 Aris, 152 Arnold, 152, 283, 397 Aronson, 415
Asano, 153, 625, 633, 636, 645 Associated Laguerre function, 706, 710, 711 Associated Laguerre polynomial, 711 Associated Legendre function, 486, 703, 705 Associated SL problem, 92, 94 Associative, 730, 752 Asymptotically stable, 401 Average value, 730, 757 Averaged conservation equation, 589 Averaged Hamiltonian, 379 Averaged Lagrangian, 592, 594 Averaged variational principle, 593 Axisymmetric acoustic radiation problem, 65 Axisymmetric biharmonic equation, 67 Axisymmetric Cauchy–Poisson problem, 69 Axisymmetric diffusion equation, 64, 135 Axisymmetric Dirichlet problem, 141 Axisymmetric initial value problem, 133 Axisymmetric wave equation, 62, 79 Bäcklund transformations, 494–498, 530, 605, 606 Balmforth, 155 Barenblatt, 416–418 Barone, 151 Ben-Jacob, 155 Benguria, 374 Benjamin, 153, 173, 527, 547 Benjamin–Bona–Mahony (BBM) equation, 154, 357, 528 Benjamin–Feir (modulational) instability, 552, 554 Benjamin–Feir side-band instability, 674
L. Debnath, Nonlinear Partial Differential Equations for Scientists and Engineers, c Springer Science+Business Media, LLC 2012 DOI 10.1007/978-0-8176-8265-1,
848
Index
Benjamin–Ono (BO) equation, 153 Benney, 152, 153, 550 Benton, 416 Berezin, 462, 463 Bernoulli equation, 175, 182 Bernoulli–Euler equation, 137 Bespalov, 153 Bessel, 85, 105, 110, 697 Bessel function, 61, 105, 106, 697 Bessel’s equation, 85, 105, 110, 698 Bessel’s inequality, 91, 728 Beta distribution, 697 Beta function, 692, 694 Bhatnagar, 636 Bianchi theorem, 497 Biharmonic equation, 10, 67, 126, 133 Bilinear expansion of Green’s function, 99 Birkhoff, 404 Birkhoff theorem, 546 Blackstock, 416 Blasius problem, 131, 142 Bleakney, 321 Bohr–Sommerfeld rule, 493 Boltzmann constant, 438 Boltzmann distribution, 438, 638 Boltzmann nonlinear diffusion problem, 406 Boltzmann transformation, 408 Bona, 154, 357, 508, 528 Bond number, 187, 453 Bore, 328 Born–Infeld equation, 193 Bound state, 478, 605 Boundary-value problem, 6, 36, 51, 61, 96, 118, 134, 141, 142, 180, 394, 402, 417, 633, 739 Boussinesq, 425, 427, 436 Boussinesq equation, 10, 152, 194, 357, 428, 434–436, 451, 452, 467, 513, 651 Brauner, 416 Breakers, 299 Breather solution, 602, 604 Bright soliton, 543 Brillouin, 569 Britton, 417 Brunt–Väisälä frequency, 358 Buckley–Leverett equation, 281 Burgers, 151 Burgers equation, 151, 296, 382, 630, 637, 638
Burgers–Huxley equation, 155 Burton, 321 Camassa, 518 Camassa and Holm (CH) equation, 157, 518, 522 Canonical form, 12 Canonical transformation, 241 Canosa, 152, 397, 417 Capillary solitary wave, 456 Capillary waves, 190 Case, 416 Cauchy conditions, 6 Cauchy data, 210, 214, 331 Cauchy integral formula, 765 Cauchy principal value, 153 Cauchy problem, 6, 7, 17, 18, 36, 42, 49, 50, 52, 126, 134, 209, 210, 222–224, 234, 254–256, 274 Cauchy–Euler equation, 84 Cauchy–Poisson problem, 58, 68, 128, 135 Cauchy–Riemann equations, 495 Caustic, 569 Cavalli-Sforva, 152 Chaotic solutions, 346 Characteristic base curve, 207 Characteristic curves, 12, 207, 316, 435 Characteristic direction, 207, 229 Characteristic equation, 12, 207, 230 Characteristic function, 755 Characteristic shock, 341 Characteristic strip, 231 Characteristic strip equations, 231 Characteristic velocity, 370 Characteristics, 12, 207, 327 Charpit equations, 230–235 Chebyshev polynomial, 708, 709 Chen, 152, 155 Chézy formula, 298 Chiu, 416 Chromatographic models, 300 Chu, 153, 378, 549 Circular frequency, 24 Clairaut equation, 233, 237 Classical solution, 3 Claude, 517 Cnoidal wave, 458, 461, 462 Cole, 386, 396 Cole–Hopf transformation, 387
Index Collins, 618 Column isotherm, 300 Commutative, 36, 730, 752 Commutator, 499 Compact breather, 514 Compactons, 509–511, 513, 514 Complementary error function, 693 Complete elliptic integral, 461, 715, 717 Complete integral, 203 Complete set, 89 Complete solution, 203 Complex amplitude, 360 Composition, 751 Compression wave, 262, 302 Concentric KdV equation, 155 Conjugate, 751 Conjugate variables, 240 Conservation law, 277, 374, 469–473, 547, 621, 670, 671 Conservation of energy, 164, 165, 192, 480 Constant of motion, 469 Continuity, 730 Continuous spectrum, 51, 479 Convolution, 36, 521, 730, 762 Convolution theorem, 36, 752, 760 Couette flow, 121 Courant, 336 Cramer, 637 Crandall, 491, 529 Crank, 409, 415 Crighton, 416 Cylindrical KdV equation, 155, 529 Cylindrical wave equation, 103 D’Alembert solution, 18, 37, 332, 434, 758 Damped harmonic oscillator, 738 Darcy law, 347 Dark soliton, 543, 575 Davey, 153, 156, 550 Davey–Stewartson equation, 156, 663, 668, 670 Davis, 153, 612 De Broglie wave, 45, 358, 577 De Leonardis, 604 De Vries, 152, 428 Debnath, 52, 55, 58, 60, 62, 68, 69, 83, 88, 115, 119–124, 128, 129, 132, 135, 142, 152, 173, 182–185, 324, 369,
849
371, 432, 460, 464, 539, 541, 542, 580, 584, 598, 622 Debye, 428 Debye length, 439 Deep water waves, 178 Defocusing branch, 509 Degasperis and Procesi (DP) equation, 157, 518, 525 Delta function, 746 Density, 284, 469 Depassier, 374 Derivative expansion method, 650 Deterministic chaos, 342, 345 Developable surface, 238 Dielectric constants, 567 Differential equation for Airy function, 702 associated Laguerre function, 711 associated Legendre function, 705 Bessel function, 697, 698 Chebyshev polynomials, 708, 710 Gegenbauer polynomials, 707 Hermite function, 713 Hermite polynomials, 712 Jacobi polynomials, 707 Laguerre polynomials, 711 Legendre polynomial, 704 modified Bessel function, 700 Weber-Hermite function, 713 Diffusion coefficient, 306 Diffusion equation, 8, 27, 29, 42, 43, 48, 49, 54, 55, 64, 71–73, 109, 112, 114, 117, 122, 132, 137, 382, 387 Diffusion kernel function, 757 Diffusion length, 348 Diffusive wave, 383, 386 Dipole solution, 393 Dirac, 381, 535, 623, 740 Dirac delta function, 34, 36, 60, 70, 740, 741, 746 Dirichlet condition, 6 Dirichlet kernel, 731, 732 Dirichlet problem, 31, 39, 47, 139, 141, 163 Discontinuous initial data, 267, 274 Discontinuous solution, 261, 270, 275–277 Dispersion relation, 45, 69, 177, 186, 187, 246, 355–359, 370, 439, 444, 569, 595, 631, 637, 661, 672 for Alfvén gravity wave, 358
850
Index
for cnoidal wave, 461 for de Broglie wave, 45 for electromagnetic waves, 358 for ion-acoustic wave, 439 for water waves, 69, 178, 357, 426, 540 Dispersion relation for de Broglie wave, 45, 358 Dispersion relation for the gravity-capillary waves, 187 Dispersion relation for water waves, 69, 178, 357, 426, 540 Dispersive wave, 357 Dissipative wave equation, 622 Distribution, 740, 747 Distributive, 730, 753 Dodd, 503, 504 Domain of dependence, 18 Drazin, 489, 492, 601 Dressler, 310, 311 Duality, 751, 761 Duhamel formula, 145, 761 Dunbar, 417 Duncan et al., 618 Dutta, 460, 464, 539, 541, 598 Dynamical equation, 587 Effective mode, 546 Eigenfunctions, 24, 28, 30, 83, 84, 112 Eigenvalues, 22, 23, 28, 30, 83, 84, 316, 492 Eikonal equation, 247, 248, 252, 569 Eilbeck, 618 Einstein, 149, 425 Ekman layer, 132 Elastic particles, 604 Electrical susceptibility, 568 Electromagnetic wave equation, 10 Electron plasma frequency, 439 Elliptic equation, 13, 14 Elliptic functions, 616, 715 Elliptic integral, 460, 539, 715 Energy, 26, 45, 101, 107, 191, 755 Energy density, 364, 366 Energy equation, 192, 197 Energy flux, 192, 364, 366 Energy inequality, 107 Energy integral, 101, 140 Engelbrecht, 636 Entropy criterion, 277 Envelope soliton, 430, 542, 543
Enz, 151 EPDiff equation, 524 Equation Ablowitz and Ladik (AL), 199 associated Legendre, 486 averaged conservation, 589 axisymmetric wave, 62 Benjamin–Bona–Mahony (BBM), 154, 357, 528 Benjamin–Ono (BO), 153 Bernoulli, 175, 182 Bernoulli–Euler, 137 Bessel, 85, 105, 110, 698 biharmonic, 10, 67, 126, 133 Born–Infeld, 193 Boussinesq, 10, 152, 194, 357, 428, 434–436, 451, 452, 467, 513, 651 Buckley–Leverett, 281 Burgers, 151, 296, 382, 630, 637, 638 Burgers–Huxley, 155 Camassa and Holm (CH), 157, 518, 522 Cauchy–Euler, 84 Cauchy–Riemann, 495 characteristic, 12, 207, 230 characteristic strip, 231 Charpit, 230–235 Clairaut, 233, 237 concentric KdV, 155 continuity, 284, 307, 438, 445 cylindrical KdV, 155, 529 Davey–Stewartson, 156, 663, 668, 670 Degasperis and Procesi (DP), 157, 518, 525 diffusion, 8, 27, 29, 42, 43, 48, 49, 54, 55, 64, 71–73, 109, 112, 114, 117, 122, 132, 137, 382, 387 dynamical, 587 eikonal, 247, 248, 252, 569 electromagnetic wave, 10 elliptic, 13, 14 Euler, 173, 174, 183, 194, 440, 446, 447, 451 Euler–Darboux, 20 Euler–Lagrange, 159, 160, 162, 168, 171, 194, 195, 197 Euler–Poincaré (EP), 524 Euler–Poisson–Darboux, 324, 351 evolution, 499, 608, 624, 627 Fisher, 152, 396
Index Fokker–Planck, 226 fractional diffusion, 114, 118, 122 fractional partial differential, 114 fractional Schrödinger, 124 fractional wave, 116, 118, 122, 123 Gardner, 471 Gelfand–Levitan–Marchenko (GLM), 483 generalized NLS, 564, 645 Ginzburg–Landau (GL), 154 Hamilton, 163 Hamilton–Jacobi, 238–240, 243, 244 Hamilton’s canonical, 524 Harry Dym, 525 heat, 8, 53 Helmholtz, 9, 65, 75, 127, 144 Hermite differential, 712 Hodgkin and Huxley, 155 hyperbolic, 12, 322, 323, 351, 370 hypergeometric, 323 internal wave, 358 internal–inertial wave, 358 inviscid derivative Burgers, 521 Johnson, 155, 447, 449 Kadomtsev–Petviashvili (KP), 155, 444, 445, 473, 529 Kawahara, 376 KdV–Burgers, 372, 508, 624, 637 kinematic wave, 151, 259, 284 Klein–Gordon (KG), 9, 126, 138–140, 151, 171, 357, 579, 582, 584, 585, 592, 619, 622 Korteweg–de Vries (KdV), 10, 38, 152, 357, 428, 438, 440, 443, 452, 458, 468, 475, 505, 530, 531, 638, 658, 660 KPP, 398 Kuramoto–Sivashinsky, 376 Lagrange, 163, 164 Laplace, 8, 9, 39, 41, 47, 59, 138, 146, 663 Lax, 499, 501, 530, 531 linear partial differential, 1–3, 10 linear scattering, 608 linear wave, 4, 8, 17, 22, 36, 49, 50, 52, 58, 62, 78–80, 102, 109, 441, 444 Lorenz, 343 Maxwell, 438, 567, 613 minimal surface, 163 modified KdV (mKdV), 153, 505, 528, 530, 635
851
Navier–Stokes, 307 nonlinear diffusion, 382, 406, 408, 409 nonlinear evolution, 500, 502 nonlinear Klein–Gordon (KG), 151, 554 nonlinear partial differential, 150, 203, 362, 650 nonlinear reaction–diffusion, 412 nonlinear Schrödinger (NLS), 153, 378, 503, 505, 529, 550, 552, 555, 558, 562, 574, 577, 654, 661, 663, 668, 671 nonlinear wave, 151, 328, 350 Otto–Sudan–Ostrovsky, 508 Painlevé, 467 parabolic, 13, 14, 383 paraxial, 571 Peregrine, 640 Poisson, 9, 73, 195 Prandtl–Blasius, 419, 420 quasi-linear partial differential, 3, 194, 202, 378 reaction–diffusion, 397, 412 Riccati, 475 roll wave, 309 Rossby wave, 130, 358 scattering, 608 Schrödinger, 9, 44, 124, 195, 253, 477, 486, 504, 507, 541, 544, 564 shallow water, 327, 348, 433 sine-Gordon (SG), 151, 506, 594, 595, 598, 605, 620, 621 sinh-Gordon, 506, 621 telegraph, 9, 126, 130, 132, 144, 145, 349 Toda lattice, 157, 158 traffic flow, 286, 287 Tricomi, 14, 351 water wave, 173, 184, 185, 432, 434 wave, 4, 17, 22, 36, 49, 50, 52, 58, 62, 78–80, 92, 101–104, 115, 122, 140, 151, 167–170, 444, 613 Whitham, 157, 193, 364, 372, 375–378, 531, 549, 593 Yang–Mills, 158 Zakharov, 562 Error function, 693 Estabrook, 495, 497, 503 Euler constant, 690 Euler equation, 173, 174, 183, 194, 440, 446, 447, 451 Euler–Darboux equation, 20
852
Index
Euler–Fourier formulas, 720, 725 Euler–Lagrange equation, 159, 160, 162, 168, 171, 194, 195, 197 Euler–Lagrange variational problem, 159, 160 Euler–Poincaré (EP) equation, 524 Euler–Poisson–Darboux equation, 324, 351 Evolution equation, 499, 608, 624, 627 Expansive wave, 261, 275 Factorial function, 689 Fan, 291, 301 Fan centered, 291, 292 Feigenbaum constant, 345, 346 Feir, 371, 547 Ferguson, 546 Fermi, 428, 547 Fermi–Pasta–Ulam recurrence, 428, 547 Fife, 152, 415, 417 Finite Hankel transform, 123, 686 Fisher, 152, 396 Fisher equation, 152, 396 Fission, 490 FitzHugh, 155 Flesch, 618 Flood wave, 297–299 Flux, 284, 469 Flux transport, 412 Focusing branch, 509 Focusing instability, 574 Fokker–Planck equation, 226 Forcing term, 356 Fordy, 504 Fourier, 381 Fourier cosine transforms, 679 Fourier integral formula, 749 Fourier integral solution, 49 Fourier method, 5, 22 Fourier series, 719–731, 733, 736, 739 Fourier sine transforms, 677 Fourier transform, 34, 46, 360, 675, 751 Fourier–Bessel series, 111 Fractional Blasius problem, 142 Fractional derivative, 55, 764 Fractional diffusion equation, 114, 118, 122 Fractional partial differential equations, 114 Fractional Rayleigh problem, 119 Fractional Schrödinger equation, 124 Fractional Stokes problem, 119
Fractional Stokes–Ekman problem, 142 Fractional unsteady Couette flow, 121 Fractional wave equation, 116, 118, 122, 123 Freeman, 155, 498, 529 Frenkel, 151 Fresnel integrals, 693 Friedrichs, 336 Froude number, 187, 310 Fulton, 153 Function Airy, 10, 38, 156, 468, 702 associated Laguerre, 706, 710, 711 associated Legendre, 486, 703, 705 Bessel, 61, 105, 106, 697 beta, 692, 694 characteristic, 755 complementary error, 693 diffusion kernel, 757 Dirac delta, 34, 36, 740, 741 eigen, 24 error, 693 factorial, 689 gamma, 689 Gaussian, 742 generalized, 740, 744, 747 good, 743 Hamilton, 163, 243 Hamilton principle, 238 Heaviside, 745 Hermite, 713 hypergeometric, 323 incomplete gamma, 692 Jacobi Epsilon, 616 Jacobi theta, 395 Laguerre, 710, 711 Legendre, 703 Mittag-Leffler, 114, 713, 766 modified Bessel, 700 normal probability density, 696 ordinary Gaussian, 742 phase, 360 Riemann, 79, 322 sawtooth wave, 723 smooth, 204 special, 689 square wave, 723, 759 theta, 464 triangular wave, 390, 722, 724
Index Weber–Hermite, 712, 713 Wright, 115 Functional, 159–162 Fundamental harmonic, 25 Fundamental parameters, 183, 438 Fundamental period, 25 Fundamental solution, 42, 43, 389, 408 Galvin, 430 Gamma distribution, 697 Gamma function, 689 Gardner, 430, 472, 474, 635 Gardner equation, 471 Gardner transformation, 471 Gardner–Morikawa transformation, 628, 633, 634 Gauss formula, 696 Gaussian function, 742 Gazdag, 417 Gegenbauer polynomial, 707, 708 Gelfand–Levitan–Marchenko (GLM) equation, 483 General integral, 204 General Parseval’s relation, 754 General solution, 4, 204 Generalized Fourier coefficients, 89, 93 Generalized Fourier series, 89 Generalized function, 740, 744, 747 Generalized Mittag-Leffler function, 713 Generalized NLS equation, 564, 645 Generalized Rankine–Hugoniot shock condition, 340 Generalized Riemann invariants, 339 Generalized simple wave, 338–340 Generalized solitary wave, 457, 458 Generalized solution, 3, 257, 271, 272, 275 Generating function for associated Laguerre function, 711 associated Legendre function, 705 Bessel function, 698 Chebyshev polynomials, 708 Gegenbauer polynomials, 707 Hermite polynomials, 712 Jacobi polynomials, 707 Laguerre function, 711 Legendre polynomial, 704 Geometrical optics approximation, 585 Ghez, 415 Gibbon, 151, 503, 563
853
Gibbs, 733 Gibbs phenomenon, 733 Ginzburg–Landau (GL) equation, 154 Glacier flow, 305 Goldstein, 243 Good function, 743 Gordon, 9 Gorschkov, 649 Gorter–Mellink law, 411 Gravity–Capillary surface waves, 187, 452 Green function, 42, 43, 70–74, 76–82, 97, 137 Green signal problem, 292 Gregory, 734 Griffith, 321 Grimshaw, 527 Grindrod, 417 Group velocity, 45, 178, 189, 357, 662 Growth rate, 396 Growth term, 259 Gurney, 415 Gurtin, 415 Haberman, 152 Hadamard example, 7 Hagstrom, 417 Hahn, 612 Haight, 287 Hamilton equations, 163 Hamilton function, 163, 243 Hamilton principle, 165, 168 Hamilton principle function, 238 Hamilton variational principle, 649 Hamilton–Jacobi equation, 238–240, 243, 244 Hamiltonian, 44, 163, 173, 375, 524 Hamiltonian density, 604 Hamilton’s canonical equation, 524 Hammack, 542 Hankel transform, 61, 446, 683 Harry Dym equation, 525 Hasegawa, 570 Hasimoto, 555 Hassan, 650 Heat equation, 8, 53 Heaviside function, 745 Heisenberg, 149, 535 Helmholtz equation, 9, 65, 75, 127, 144
854
Index
Hermite differential equation, 712 Hermite function, 713 Hermite polynomial, 712 Hilbert, 201, 355, 579, 623 Hilbert transform, 153 Hirota, 430, 617 Hodgkin and Huxley equation, 155 Hodograph transformation, 336 Holm, 522 Hopf, 386 Hoppensteadt, 402 Howard, 413 Hydraulic jump, 299, 328 Hyman, 509–511, 513, 515 Hyperbolic equation, 12, 322, 323, 351, 370 Hypergeometric equation, 323 Hypergeometric function, 323 Ichikawa, 153 Identity, 36, 753 Ilichev, 376 Ill posed, 7, 146, 147, 370 Incomplete gamma function, 692 Index of refraction, 248 Infeld, 563 Initial value problem, 17, 128, 133, 219, 259, 313, 331, 359 Inner product, 87, 89, 755 Inoue, 655 Instantons, 158 Integrability condition, 230 Integral equation, 765 Integral surface, 4, 206 Interaction strength, 546 Internal wave equation, 358 Internal–inertial wave equation, 358 Inverse Fourier transform, 34, 46 Inverse Hankel transform, 61 Inverse Laplace transform, 51 Inverse scattering transform, 475, 476, 478 Inverse Weierstrass transform, 763 Inviscid derivative Burgers equation, 521 Ion plasma frequency, 439 Ion-acoustic wave, 438, 440, 638 Isentropic flow, 314, 351 Isothermal curves, 40 Jacobi dn function, 541, 716 Jacobi sn function, 460, 716
Jacobi elliptic function, 461, 616, 716 Jacobi elliptic integrals, 715 Jacobi Epsilon function, 616 Jacobi polynomial, 706, 707 Jacobi principle of least action, 245, 248 Jacobi theta function, 395 Jeffrey, 508, 569, 636, 655 Johnson, 417, 445, 489, 661 Johnson equation, 155, 447, 449 Jones, 360, 743 Josephson, 151 Jost solutions, 481 Jump condition, 257, 270, 274 Jump discontinuity, 270 Kadomtsev, 155 Kadomtsev–Petviashvili (KP) equation, 155, 444, 445, 473, 529 Kako, 152 Kakutani, 153, 625, 636 Kaliappan, 413 Kametaka, 417 Karpman, 153, 462, 509, 563 Kaup, 506, 509 Kawahara, 636, 650, 655 Kawahara equation, 376 KdV–Burgers equation, 372, 508, 624, 637 Keller, 417 Kelley, 153 Kerr effect, 570 Kinematic wave equation, 151, 259, 284 Kinetic energy, 26, 45, 163, 164, 167, 190, 196, 524 Kink, 431 Kirchhoff transformation, 409 Kivshar, 515, 516 Klein, 9 Klein–Gordon (KG) equation, 9, 126, 138–140, 151, 171, 357, 579, 582, 584, 585, 592, 619, 622 Kleitz–Seddon law, 299 Kodama transformation, 522 Kolmogorov, 152, 397 Konno, 577 Kontorova, 151 Kopell, 413 Korteweg, 152, 428 Korteweg–de Vries (KdV) equation, 10, 38, 152, 357, 428, 438, 440, 443, 452,
Index 458, 468, 475, 505, 529–531, 638, 658, 660 KPP equation, 398 Kramers, 569 Kriess, 416 Kronecker delta, 24, 240 Krushkal, 153 Kruskal, 428, 429, 525, 547 Kulikovskii, 636 Kuramoto–Sivashinsky equation, 376 Kynch, 151, 303, 348 Lagrange bracket, 242, 243 Lagrange equation, 163, 164 Lagrange identity, 86 Lagrangian, 163, 168, 172, 198, 367 Lagrangian density, 168, 604, 621 Lagrangian functional, 169 Lagrangian stability, 546 Laguerre differential equation, 711 Laguerre function, 710, 711 Laguerre polynomial, 710, 711 Lake, 153, 547 Lamb, 151, 612 Lamé constants, 127 Landau, 493 Langmuir isotherm model, 301 Langmuir soliton, 559 Laplace equation, 8, 9, 39, 41, 47, 59, 138, 146, 663 Laplace spherical harmonics, 105 Laplace transform, 51, 59, 680, 758 Laplacian, 44, 73, 79, 102, 109, 742 Lardner, 416 Larson, 416, 417 Lattice solitary waves, 618 Law of conservation of energy, 164, 165, 480 Lax, 257, 277, 340, 811 Lax entropy criterion, 277 Lax equation, 499, 501, 530, 531 Lax pair, 499, 530 Legendre duplication formula, 691 Legendre functions, 703 Legendre polynomial, 487, 703, 704, 708 Leibniz, 734 Leibniz series, 734 Leibovich, 153 LeVeque, 341
855
Levi-Civita, 426 Lie group, 404, 466 Lifshitz, 493 Lighthill, 151, 152, 194, 283, 286, 369, 416 Lighthill criterion, 566 Linear operator, 3, 70 Linear partial differential equation, 1–3, 10, 105 Linear scattering equation, 608 Linear superposition principle, 5, 6, 43, 74 Linear wave equation, 4, 8, 17, 22, 36, 49, 50, 52, 58, 62, 78–80, 102, 109, 441, 444 Liu, 152, 155 Local frequency, 361, 362, 367 Local Mach number, 335 Local wavenumber, 361, 362, 367 Localized modes, 516 Logan, 417 Logistic map, 345 Lommel integrals, 699 Long wavelength parameter, 183 Lorenz, 342, 417 Lorenz attractor, 342 Lorenz equation, 343 Lorenz system, 342 Luke, 171 Ma, 576 MacCamy, 415 Mach lines, 335 Mach number, 252, 335 Madsen, 152 Magnetic permeability, 567 Mainardi, 115, 118 Makhankov, 430 Manning formula, 298 Manoranjan, 417 Marchenko, 376 Matrix operators, 502 Matsumoto, 655 Maximum principle, 108 Maxon, 156 Maxwell equation, 438, 567, 613 Maxworthy, 430 McCall, 612 McKean, 417 Mean value property, 34 Mei, 153, 378, 549
856
Index
Mersenne law, 25 Method of characteristics, 208, 333 Method of multiple scales, 656, 661 Minimal surface equation, 163 Mitchell, 417 Mittag-Leffler function, 114, 713, 766 Miura, 470, 475, 528 Miura transformation, 470 Modes, 25 Modified Bessel equation, 700 Modified Bessel function, 701 Modified KdV (mKdV) equation, 153, 505, 528, 530, 635 Modulational instability, 554, 564 Mollenauer, 570 Moment of instability, 436 Momentum equation, 312 Monge axis, 207 Monge cone, 208, 229 Monochromatic waves, 568 Monoclinal flood wave, 299 Montroll, 413 Morikawa, 628 Multiple Fourier transform, 46, 47 Munier, 409 Murray, 413–417 N -soliton solution, 486, 762 n-th convolution product, 762 N -wave solution, 392 Nagumo, 155 Naumkin, 376, 377 Navier–Stokes equation, 307 Nayfeh, 153 Neumann boundary-value problem, 6 Neumann conditions, 6 Newell, 604 Newman, 413–416 Nigmatullin, 115 Nisbet, 415 Nishihara, 637 Nodes, 25 Non-periodic solutions, 738 Nondispersive wave, 178, 357, 452 Nonlinear diffusion equation, 382, 406, 408, 409 Nonlinear Dirichlet problem, 163 Nonlinear dispersion relation, 181, 516, 540, 573, 620
Nonlinear evolution equation, 500, 502 Nonlinear group velocities, 370 Nonlinear Klein–Gordon (KG) equation, 151, 554 Nonlinear lattices, 615 Nonlinear partial differential equation, 150, 203, 362, 650 Nonlinear reaction–diffusion equation, 412 Nonlinear Schrödinger (NLS) equation, 153, 378, 503, 505, 529, 550, 552, 555, 558, 562, 574, 577, 654, 661, 663, 668, 671 Nonlinear shallow water waves, 354 Nonlinear superposition principle, 497, 607 Nonlinear wave equation, 151, 328, 350 Normal dispersion, 575 Normal modes, 24 Normal probability density function, 696 Novick-Cohen, 376 Nozaki, 636 Numerical series, 734, 736, 737 Nwogu, 152 Nye, 151, 305 Oikawa, 636 Okubo, 408 Olver, 173 Ono, 153, 550, 625, 633, 636 Order, 2 Ordinary Gaussian functions, 742 Ordinary output, 762 Orthogonal relation, 706, 707, 711 Orthogonal set, 89 Ostrovsky, 508, 624, 646 Otto–Sudan–Ostrovsky equation, 508 Outi, 542 Overtones, 25 p-mode, 517 Pack, 321 Page, 517 Painlevé equation, 467 Parabolic equation, 13, 14, 382, 383 Paraxial equation, 571 Parity relation, 709, 712 Parker, 416 Parseval formula, 727 Parseval relation, 91, 729, 755 Partial differential equation, 2
Index
857
Particular solution, 204 Pasta, 428, 547 Pattle, 413 Pawlik, 153 Peakon solitary traveling wave, 523 Peakons, 519, 522, 524 Peierls–Nabarro potential, 517 Pelinovsky, 624, 646 Penel, 416 Peregrine, 152, 527, 576 Peregrine equation, 640 Period-doubling, 346 Perring, 600, 608 Petviashvili, 155 Phase function, 360 Phase velocity, 177, 245, 661 Phillips, 577 Piston problem, 319, 325 Planck constant, 9, 44, 124, 358, 494 Plane waves, 102 Plateau problem, 162, 197 Platzman, 416 Poincaré, 1, 227 Poincaré return map, 345 Poincaré theorem, 241 Point of inflection, 414 Pointwise Convergence Theorem, 732 Poisson bracket, 240 Poisson equation, 9, 73, 195 Poisson integral formula, 34, 39, 43 Poisson kernel, 34 Poisson summation formula, 752 Polarization, 568 Polytropic gas, 319 Potential, 507, 550, 608 Potential energy, 26, 45, 165, 168, 190, 196 Power spectrum, 755 Prandtl–Blasius equation, 419, 420 Prasad, 152, 153, 636 Principle of least action, 198 Prolongation structure method, 503 Properties of the convolution, 756
Rankine–Hugoniot condition, 270, 340 Ravindran, 152, 153 Rayleigh, 427 Rayleigh formula, 190 Rayleigh problems, 119 Rayleigh quotient, 545 Rayleigh–Benard convection problem, 344 Reaction term, 259, 397 Reaction–diffusion equation, 397, 412 Recurrence relation, 699, 706, 711, 712 Redekopp, 152 Reductive perturbation method, 627 Reflected wave, 479 Reflection coefficient, 479 Refractive index, 246, 568, 570 Refractive wave, 261, 275 Resonance, 738 Reynolds number, 386, 388, 393, 395 Riabouchinsky, 434 Riccati equation, 475 Richards, 151, 286 Riemann, 311, 385 Riemann function, 79, 322 Riemann initial-value problem, 313 Riemann invariants, 312, 317–320, 325, 329, 330 Riemann–Lebesgue Lemma, 728, 751 Riemann–Liouville fractional integral, 764 Ripples, 189 Robin boundary-value problem, 6 Robin condition, 6 Rodin, 416 Rodrigues formula, 703, 707, 709–712 Roll wave equation, 309 Rosales, 468 Rosenau, 417, 509, 511, 513, 515, 518 Roskes, 153, 550 Rossby wave equation, 130, 358 Rowland, 563 Rowlands, 153, 563 Ruled surface, 238 Russell, 425, 426
Quasi-linear partial differential equation, 3, 194, 202, 378
s-mode, 517 Saddle point, 401 Saffman, 526 Sagdeev, 440 Sandri, 650 Sandusky, 517
Radian frequency, 24 Raizer, 416 Rankine, 182, 371
858
Index
Sanuki, 564 Saric, 153 Satsuma, 155 Sawtooth wave function, 723 Scaling, 751, 759 Scattering data, 475 Scattering equation, 608 Scattering state, 476, 479 Schiff, 151 Schimizu, 153 Schmidt, 563 Schneider, 115 Schrödinger equation, 9, 44, 124, 195, 253, 477, 486, 504, 507, 541, 544, 564, 670 Schwartz space, 755 Schwarz inequality, 545 Scott, 151, 155, 416, 619 Secular (unbounded) term, 652 Seddon, 299 Segur, 155, 425, 488, 506 Self-focusing instability, 570 Self-induced transparency (SIT), 612 Seliger, 376 Sen, 136, 637 Separation of variables, 21, 22, 29, 32, 598 Separatrix, 401 Serret–Frenet formulas, 556 Shabat, 502, 535, 542, 544, 545, 547, 550, 576 Shallow water equation, 327, 348, 433 Shallow water speed, 431, 459 Shallow water waves, 178, 327 Shifting, 751, 759 Shigesada, 415 Shishmarev, 376, 377 Shock condition, 270, 271, 340 Shock wave, 205, 264, 270, 288, 321 Sievers, 516, 517 Sign function, 745 Similarity solution, 404, 406, 409 Similarity transformation, 406 Similarity variable, 389, 406 Simple harmonic oscillator, 244, 737 Simple wave, 311, 314, 318 Sine-Gordon (SG) equation, 151, 506, 594, 595, 598, 605, 620, 621 Single-soliton solution, 604, 610 Singular solution, 204, 249 Sinh-Gordon equation, 506, 621
Skyrme, 151 Sleeman, 155 Slobodkina, 636 Smoller, 341, 417 Smooth function, 204 sn function, 460 Solitary wave, 429, 430, 436, 457, 540–542, 600 Solution complete, 203 discontinuous, 261, 275–277 general, 4, 204 generalized, 3, 257, 271, 272, 275 particular, 204 singular, 204, 249 weak, 3, 257, 271–274, 341 Sorensen, 152 Source solution, 389 Source term, 259, 397 Sparrow, 420 Special functions, 689 Spectrum, 83 Speed of shallow water waves, 178, 428, 431, 434 Spherical waves, 103 Square wave function, 723, 759 Stable node, 400 Standing wave, 25 Stationary phase method, 60, 69, 391 Stationary point, 60, 67, 360, 391 Stewartson, 153, 154, 156, 550 Stirling approximation, 692 Stokes, 355, 426 Stokes’ analysis, 182 Stokes expansion, 182, 369 Stokes problem, 119, 131 Stokes wave, 179, 552 Stokes–Ekman problem, 131, 142 Strange attractor, 345 Stretched variable, 632 Stretching transformation, 404 Strong convergence, 90 Stuart, 154 Sturm–Liouville systems, 82–85, 138, 139, 501 Sturrock, 650 Su, 635
Index Superposition principle linear, 5, 6 nonlinear, 497 Supersonic flow, 334, 341 Surface area of a sphere in n-dimension, 690 Surface wave problem, 128, 134 Symmetric, 692 Takeno, 516, 517 Talanov, 153 Tang, 417 Taniuti, 153, 625 Tappert, 153 Telegraph equation, 9, 126, 130, 132, 144, 145, 349 Thermal conductivity, 54 Theta function, 464 Three-soliton solution, 488 Thyagaraja, 544 Toda, 152, 158, 615 Toda lattice equation, 157, 158 Toda lattice soliton, 158, 615, 618 Toland, 182 Traffic flow, 286, 287 Traffic flow equation, 286, 287 Traffic signal problem, 294 Transform finite Hankel, 686 Fourier, 34, 46, 360, 675, 751 Fourier cosine, 679 Fourier sine, 677 Hankel, 61, 446, 683 Hilbert, 153 inverse scattering, 475, 476 Laplace, 51, 59, 680, 762 Weierstrass, 763 Transition layer, 386 Transition region, 271 Translation, 751 Transmission coefficient, 479 Transmitted wave, 479 Transverse vibration of string, 22 Traveling wave solution, 50, 384 Triangular wave function, 390, 722, 724 Tricomi equation, 14, 351, 354 Trochoid, 180 Trullinger, 604 Turin, 412
859
Two-sided Laplace transform, 762 Two-soliton solution, 488, 497, 498 Ulam, 428, 547 Undular bore, 461 Unitary transformation, 755 Van Saarloos, 155 Van Wijngaarden, 153 Vanden-Broeck, 527 Variational principle, 159, 171, 172, 193, 196 Varma, 153 Velocity of sound, 312 Velocity potential, 175 Viecelli, 156 Volume of a sphere in n-dimension, 690 Von Neumann, 149, 201, 227 Vortex filament, 555 Vorticity, 174 Wadati, 152, 577 Wahlquist, 495, 497, 503 Walsh, 396, 416 Wang, 155 Ward, 158 Washimi, 153, 625 Water wave equation, 173, 184, 185, 432, 434 Watson, 650 Wave Alfvén gravity, 358 capillary solitary, 456 cnoidal, 458, 461 compression, 262, 302 de Broglie, 45, 358, 577 diffusive, 383, 386 dispersive, 357 electromagnetic, 10 expansive, 261, 275 flood, 297, 298 generalized simple, 338–340 generalized solitary, 457, 458 gravity, 358 gravity–capillary, 187, 452 incident, 486 inertial, 358 internal, 358 internal-inertial, 358 ion-acoustic, 438, 440, 638
860
Index
kinematic, 151, 259 N , 392 nondispersive, 178, 357 reflected, 479 refractive, 261, 275 roll, 309 Rossby, 130, 358 shallow water, 178, 640 shock, 205, 264, 270, 288, 321 simple, 275, 311, 314, 318, 320 solitary, 429, 430, 436, 457, 540–542 standing, 25 Stokes, 179, 552 transmitted, 479 water, 176, 357 Wave diffraction, 572 Wave energy, 546 Wave equation, 4, 8, 17, 22, 36, 49, 50, 52, 58, 62, 78–80, 92, 101–104, 115, 122, 126, 140, 151, 167–170, 444, 613 Wave front, 245 Wave propagation velocity, 285 Wavepacket, 570 Weak solution, 3, 257, 271–274, 341, 519 Webber, 417 Webber solution, 698 Weber–Hermite function, 712, 713 Wei, 625 Weidman, 152 Weierstrass transform, 763
Weiland, 153 Well posed problem, 7, 18, 147 Wentzel, 569 West, 413 Whitham, 151, 171, 173, 193, 286, 299, 359, 371, 377 Whitham averaged Lagrangian, 368, 592 Whitham averaged variational principle, 367, 368, 378 Whitham equation, 157, 193, 364, 366, 368, 372, 375–378, 531, 549, 593 Whitham rule, 290, 392 Wigner, 579 Wiljelmsson, 153 WKBJ method, 569 Wright function, 115 Wronskian, 480 Wyss, 115 Yajima, 152 Yang–Mills equation, 158 Yuen, 153, 546 Zabusky, 152, 428–430, 462, 469, 547 Zakharov, 152, 173, 502, 535, 542, 544, 545, 547, 550, 576 Zakharov equation, 562 Zakharov–Shabat (ZS) scheme, 502 Zel’dovich, 416, 418 Zeppetella, 417 Zufira, 376